ieee communications surveys & tutorials 1 a survey on

IEEE COMMUNICATIONS SURVEYS amp TUTORIALS 1

A Survey on Mobile Crowdsensing SystemsChallenges Solutions and Opportunities

Andrea Capponi Student Member IEEE Claudio Fiandrino Member IEEE Burak Kantarci Senior Member IEEELuca Foschini Senior Member IEEE Dzmitry Kliazovich Senior Member IEEE and Pascal Bouvry Member IEEE

AbstractmdashMobile crowdsensing (MCS) has gained significantattention in recent years and has become an appealing paradigmfor urban sensing For data collection MCS systems rely oncontribution from mobile devices of a large number of partici-pants or a crowd Smartphones tablets and wearable devicesare deployed widely and already equipped with a rich setof sensors making them an excellent source of informationMobility and intelligence of humans guarantee higher coverageand better context awareness if compared to traditional sensornetworks At the same time individuals may be reluctant to sharedata for privacy concerns For this reason MCS frameworksare specifically designed to include incentive mechanisms andaddress privacy concerns Despite the growing interest in theresearch community MCS solutions need a deeper investigationand categorization on many aspects that span from sensing andcommunication to system management and data storage In thispaper we take the research on MCS a step further by presentinga survey on existing works in the domain and propose a detailedtaxonomy to shed light on the current landscape and classifyapplications methodologies and architectures Our objective isnot only to analyze and consolidate past research but also tooutline potential future research directions and synergies withother research areas

Index TermsmdashMobile crowdsensing urban sensing opportunis-tic sensing participatory sensing

I INTRODUCTION

MOBILE crowdsensing (MCS) has gained popularityin recent years becoming an appealing paradigm for

sensing and collecting data MCS systems rely on sensorsand communication interfaces embedded in commonly usedmobile devices such as smartphones and wearables Nowadaysmobile devices are essential for our daily activities includingbusiness communication and entertainment [1] [2] Accordingto Gartner statistics the number of worldwide smartphonessales in 2018 was 155 billion units [3] and the numberof wearable devices shipped in 2018 was 17891 millionwhich is projected to reach 45319 million in 2022 [4] Smartwatches glasses rings gloves and helmets are the mostpopular wearable devices currently available on the marketcorresponding to a highly increasing revenue that is estimated to

A Capponi and P Bouvry are with the University of LuxembourgLuxembourg E-mail firstnamelastnameunilu

C Fiandrino is with Imdea Networks Institute Madrid Spain E-mailclaudiofiandrinoimdeaorg

B Kantarci is with the School of Electrical Engineering and Com-puter Science University of Ottawa Ottawa ON Canada E-mail bu-rakkantarciuOttawaca

L Foschini is with DISI Univ Bologna Viale Risorgimento 2 40136Bologna Italy E-mail lucafoschiniuniboit

D Kliazovich is with ExaMotive Luxembourg E-mail kliazovichieeeorg

rise up to USD 953 billion by 2021 [5] Furthermore the crowdanalytics market is predicted to reach USD 1 1425 million by2021 raising from USD 3851 million of 2016 at a compoundannual growth rate of 243 [6]

The term mobile crowdsensing was first introduced by Gantiet al to indicate a more general paradigm [7] than mobilephone sensing [8] [9] Guo et al in [10] give a definitionthat clearly highlights this difference ldquoMCS is a new sensingparadigm that empowers ordinary citizens to contribute datasensed or generated from their mobile devices aggregates andfuses the data in the cloud for crowd intelligence extractionand people-centric service deliveryrdquo To operate efficientlyMCS systems require the participation and contribution ofa large number of users Although entire communities canpotentially benefit from such a contribution a singular personmay be reluctant to participate being selfish or having privacyconcerns To ease this burden in the last years the researchcommunity has put lots of effort in developing proper incentivemechanisms [11]ndash[14] and in investigating privacy issues [15][16]

The capillary spread of smartphones and wearables alongwith the rich set of built-in sensors are certainly the mainkey enablers leading to the success of MCS paradigmAccelerometer gyroscope GPS microphone and cameraare only a representative set of sensors that facilitated thedevelopment of several applications in a wide range ofscenarios including health care environmental and trafficmonitoring Many applications using smartphone sensors havebeen already developed and are currently in use To illustraterepresentative examples HealthAware [17] MPCS [18] andDietSense [19] foster healthy eating by collecting images ofconsumed food and inspect daily user-activity by extractingcontext information such as time and location where foodwas consumed For this purpose both applications use theaccelerometer GPS and microphone Nericell [20] monitorstraffic conditions GasMobile [21] HazeWatch [22] and Third-Eye [23] rely on active citizen participation to monitor airpollution Creekwatch [24] developed by the IBM Almadenresearch center permits to monitor the conditions of thewatershed through crowdsensed collected data about the amountof water in the river bed the amount of trash in the river bankthe flow rate and a picture of the waterway Garbage Watch [25]and WasteApp [26] allow monitoring the content of recyclingbins with the objective to improve the recycling program

MCS can significantly improve citizens everyday life andprovide new perspectives to urban societies MCS is anessential solution for building smart cities of the future which

This is the authors version of an article that has been published in this journal Changes were made to this version by the publisher prior to publicationThe final version of record is available at httpdxdoiorg101109COMST20192914030

Copyright (c) 2019 IEEE Personal use is permitted For any other purposes permission must be obtained from the IEEE by emailing pubs-permissionsieeeorg


SensingLayer

CommunicationLayer

DataLayer

ApplicationLayer

Fig 1 Layered architecture of MCS systems It shows the four-layered architecture that describes the MCS paradigm including application (Fig 5) data(Fig 6) communication (Fig 7) and sensing (Fig 8) layers which are discussed in Sec III-A

aim at using ICT solutions to improve the management ofeveryday life of their citizens [27] [28] The Internet ofThings (IoT) paradigm is the candidate solution for a widedeployment of sensing infrastructure enabling smart citiesrsquoapplications [29] Moreover active participation of citizenscan improve spatial coverage of already deployed sensingsystems with no need for further investments MCS leverageshuman intelligence which has a deeper context understandingthan traditional sensor networks Cities are facing significantdeficits in infrastructure services and humans can be involvedto improve their monitoring and maintenance We illustratethis concept with specific use cases Data harvested fromsmartphonesrsquo accelerometers over moving vehicles enablesthe detection of bridge vibrations [30] Other possible cityservices where MCS plays a fundamental role are smarttraffic management [31] [32] or free parking spot detectionSpecifically ParkSense detects vacant parking spots using WiFiscans of smartphones [33] while ParkGauge reports real-timecrowdsensed information about indoor parking occupancy andexploits low-consuming sensors (eg accelerometer) to detectthe driving states [34]

There is a wealth of literature on MCS systems Considerableresearch efforts proposed novel sensing architectures andas already mentioned investigated specific aspects such asincentive mechanisms privacy issues as well as the reliabilityand trust of the data collection process With such a large bodyof work surveys on the topic covering specific areas such asprivacy [15] or incentives for user recruitment [12] [13] already

exists Others like [35] present many aspects components ofMCS as an emerging paradigm eg data collection qualityassurance etc However this is an early work missing a wealthof early developments in the field What is missing in thispicture is a survey that covers the whole decade of research inMCS and provides a clear state of its evolution This is oneof the purposes of our manuscript Also the vast amount ofwork in literature remains uncategorized with many of thecore paradigms unclear To illustrate for example there isno consensus on the term ldquoopportunistic sensingrdquo Accordingto Ganti et al [7] ldquoopportunistic sensingrdquo is defined as ldquoOnthe other hand opportunistic sensing is where the sensingis more autonomous and user involvement is minimal (egcontinuous location sampling)rdquo However Khan et al [9] statethat opportunistic sensing requires no user involvement at allsince the decisions to perform the sensing are a prerogative ofthe device itself Finally Han et al [36] enlarge the previousvision of opportunistic sensing in the context of single-userinvolvement and they describe opportunistic sensing as aparadigm enabling cooperation among smartphones Typicallyboth terms ldquoopportunisticrdquo and ldquoparticipatoryrdquo sensing remainunder the common umbrella of MCS [7] [9] [36] [37] In othercases both ldquomobile crowdsensingrdquo and ldquoparticipatory sensingrdquoare used interchangeably [31] Other times both ldquomobilecrowdsensingrdquo ldquoparticipatory sensingrdquo and ldquoopportunisticsensingrdquo are synonyms [38]

With this plethora of definitions our manuscript has theprecise objective of proposing detailed taxonomies to sim-




Sec IIBackground

Related surveysIIA

TimelineIIB

Factors Contributingto the Rise of MCS

IIC

Acronyms used inthe MCS Domain

IID

Featureextraction

Sec IIIMCS in a nutshell

Data CollectionFrameworks

IIIB

MCS as a layeredarchitecture

IIIA

Theoretical Works OperationalResearch and Optimization

IIIC

Platforms Simulatorsand Datasets

IIID

MCS as a Business ParadigmIIIE

Final RemarksIIIF

Sec IVTaxonomies and Classification

on Application Layer

TaxonomiesIVA

ClassificationIVB

Task

User

Sec VTaxonomies and Classification

on Data Layer

TaxonomiesVA

ClassificationVB

Management

Processing

Sec VITaxonomies and Classification

on Communication Layer

TaxonomiesVIA

ClassificationVIB

Technologies

Reporting

Sec VIITaxonomies and Classification

on Sensing Layer

TaxonomiesVIIA

ClassificationVIIB

Elements

Samplingprocess

Sec VIIIDiscussion

Looking Back to See Whatrsquos NextVIIIA

Inter-disciplinary InterconnectionsVIIIB

Provides a perspective analysis of future research directions given the past efforts Exposes interconnections between MCS and other research areas

Sec IXConcluding Remarks

Fig 2 Survey organization Section II provides a background on MCS literature Section III presents the four-layered architecture and discusses theoreticaland practical works Section IV V VI and VII propose taxonomies and classification on the four layers ie application data communication and sensinglayers Section VIII discusses future directions and interconnections with other research areas Finally Section IX concludes the survey

plify the understanding of the current definitions availabletechniques and solutions in the field of MCS We foreseeto categorize MCS works in a four-layered architecture asshown in Fig 1 The underneath reason better detailed afterthat is to cover the entire process chain since the sensorproduces a reading up to when data reaches the applicationlayer The four-layered architecture is structured as followsThe top layer is the application layer which involves high-levelfunctionalities such as user recruitment and task allocation [39]The second is the data layer responsible for aspects relatedto store analyze and process collected information Thethird layer involves the communication layer comprising thecommunication technologies for the delivery of sensed dataFinally the bottom layer close to the physical layer is thesensing layer Our approach goes beyond such classification byproposing specific taxonomies for each layer of the architecture

Specifically the synopsis of contributions of the currentsurvey is as follow

bull To introduce MCS as a four-layered architecture dividedinto application data communication and sensing layers

bull To compare and analyze existing MCS data collection

frameworks (DCFs) theoretical works leveraging oper-ational research and optimization practical ones suchas platforms simulators and those making crowdsenseddatasets publicly available

bull To propose novel and detailed taxonomies based on thelayered architecture that cover all MCS aspects allowingfor a simple and clear classification of MCS systems anddomains

bull To classify MCS systems according to the proposedtaxonomies

bull To discuss future directions given the consolidated pastefforts and inter-disciplinary research areas

Survey organization Fig 2 shows the organization of thissurvey Section II provides a background on related surveys anda timeline including the most relevant works followed by themost important technological factors and related fields that havecontributed to the rise of MCS and finally lists acronyms usedin the MCS domain Section III presents MCS in a nutshellintroducing it as a four-layered architecture presenting itsmain data collection frameworks (DCFs) discussing theoretical




works on operational research and optimization problemsbut also more practical ones about platforms and simulatorsexploited for crowdsensing campaigns and collected datasetsThen it discusses MCS as a business paradigm and concludeswith final remarks summarizing previous contents and intro-ducing taxonomies based on the presented layered architectureSection IV elaborates on novel taxonomies on the applicationlayer and overviews papers accordingly Section V proposesnovel taxonomies on the data layer and surveys existingsolutions according to the presented taxonomies Section VIdiscusses novel taxonomies on the communication layer andconsequently overviews works Section VII presents noveltaxonomies on the sensing layer and classifies MCS systemsaccordingly Section VIII provides an overall discussion onthe subject by presenting a prospective analysis of futureresearch directions given the past efforts and inter-disciplinaryinterconnections with other research areas Finally Section IXconcludes the survey

II BACKGROUND

This section explores the motivations giving rise to MCS andthe reasons that make it a prominent sensing paradigm To thisend we analyze and extract from the large body of literaturethe works that can be defined as milestones ie the works thathave strongly influenced the research in the field This analysisconsiders the evolution of technologies in different domainssuch as computing and communications which significantlyaffect the efficiency of the data collection process Specificallywe overview in Section II-A related surveys to guide the readerinto past research and in Section II-B we show the temporalevolution of milestones works in the area We concludethe section by providing an overview of the main factorscontributing to the rise of MCS (Section II-C) and present howMCS services are delivered (Section II-D)

A Related Surveys

This Section presents surveys that relate with conceptsof MCS (see Table I) We categorize these works intosix categories surveys on MCS (ie those about sensingequipment) those analyzing wireless sensor networks mobilephone sensing anticipatory mobile computing user recruitmentand privacy issues

Mobile Crowdsensing Guo et al [35] coin the term MobileCrowd Sensing and Computing (MCSC) which investigates thecomplementary nature of the machine and human intelligencein sensing and computing processes Citizens contribute datawith their mobile devices to the cloud which enforces crowdintelligence Guo et al also introduce the concept of visualcrowdsensing in [40] by presenting its strengths and challengeslike multi-dimensional coverage needs low-cost transmissionand data redundancy In [42] provides a discussion on theimplementation requirements of MCS systems Like ourproposal the focus is on four dimensions task assignmentstimulation of the participation data collection and processingLiu et al survey the challenges of MCS systems and theproposed solutions for the effective use of resources [43] An

overview of different applications of MCS systems in thecontext of IoT and smart cities is provided in [44] In [41]the authors focus on citizens as data consumers and dataproducers to infer social relationships and human activitiesIn particular they overview applications on social sensorsbased on social sensor receiver platform (eg Twitter) andpresent a classification including public security smart cityand location-based services Crowdsensed data can be maliciousor unreliable Despite assessing Quality of Information (QoI)is important few works delved into the roots of the problemIn [54] the authors overview existing works assessing QoI andpropose a framework to enforce it

Sensors amp WSN Sensors are an essential component for dataacquisition Sensing equipment is typically embedded in mobiledevices but specific applications (eg air quality monitor-ing [21] [55] [56]) can employ dedicated hardware connectedwith wireless technology (eg Bluetooth) Ming presentsa study of mobile sensing [48] by providing an overviewof typical sensors embedded in smartphones and discussingrelated mobile applications The actors in a crowdsensingcampaign can be seen as nodes of a Wireless Sensor Network(WSN) In the last years developments in communicationtechnologies and electronics lead to the significant progress ofsensor networks and typically survey in the area analyze bothsensing and networking aspects [45] [47] and [46] Remotesensing technologies are discussed in [49] where the authorspresent platform and sensor developments for emerging newremote sensing satellite constellations sensor geo-referencingand supporting navigation infrastructure

Mobile Phone Sensing Mobile phone sensing can be seen asthe forefather of mobile crowdsensing This paradigm was verypopular when mobile phones did not have the capabilities ofcurrent smartphones in terms of storage communication andcomputation Unlike in MCS the research on mobile phonesensing focused mostly on individual sensing applications suchas elderly fall detection or personal well-being Several worksexploit this paradigm proposing methodologies and solutionscollecting data from sensors of mobile phones [8] [9]

Anticipatory mobile computing amp networking The richdata availability enables the possibility to infer and predictcontext and user behavior In [50] the authors overview theliterature in mobile sensing and prediction focusing on theconcept of anticipatory mobile computing The survey presentsa plethora of phenomena like user destination or behaviorthat smartphones can infer and predict by leveraging machinelearning techniques and proactive decision making In [51] theauthors analyze the concept of anticipatory mobile networkingto study pattern and periodicity of user behavior and networkdynamics The ultimate goal is to predict context and optimizenetwork performance The authors provide an in-depth overviewof the most relevant prediction and optimization techniques

User Recruitment The success of a crowdsensing campaignrelies on large participation and contribution of citizens Tothis end incentives are fundamental Existing surveys on userrecruitment investigate how to propose incentive mechanismsthat could efficiently and effectively motivate users in sensing




TABLE IRELATED SURVEYS

TOPIC DESCRIPTION REFERENCES

Mobile Crowdsensing Include works that survey crowdsensing architectures frameworks and datacollection techniques

[35] [40] [41] [42]ndash[44]

Sensors amp Sensor Networks Describe generic sensing equipment when employed by crowdsensing applica-tions sensor networks and platforms in different domains

[45]ndash[49]

Mobile Phone Sensing Describe methodologies of employment of sensing equipment embedded inmobile devices for non-crowdsensed applications

[8] [9]

Anticipatory Mobile Computing amp Networking Describe techniques like machine learning to predict the context of sensing andnetwork state

[50] [51]

User Recruitment Survey techniques to recruit users for sensing campaigns and describe existingincentive mechanisms to promote participation

[12] [13] [52] [53]

Privacy Present the threats to users and privacy mechanisms that are exploited in existingcrowdsensing applications to address these issues

[15] [16]

and reporting information In [52] the authors overview existingincentive mechanisms in participatory sensing while [53]proposes a taxonomy while providing application-specificexamples In [12] the authors analyze strategies to stimulateindividuals in participating in a sensing process They classifyresearch works into three categories namely entertainmentservice and money The first category considers methods thatstimulate participation by turning sensing tasks into gamesService-based mechanisms consist of providing services to theusers in exchange for their data Monetary incentives methodsdistribute funds to reward usersrsquo contribution

Privacy Data collection through mobile devices presents manyprivacy breaches such as tracking a userrsquos location or disclosingthe content of pictures or audio In [16] the authors providean overview of sensing applications and threats to individualprivacy when personal data is acquired and disclosed Theyanalyze how privacy aspects are addressed in existing applica-tion scenarios assessing the presented solutions and presentingcountermeasures Other works focus on privacy concerns intask management processes such as user recruitment and taskdistribution [15]

The Novelty of This Survey This survey goes beyond theworks mentioned above The focus is to categorize the literatureaccording to the corresponding ldquotechnologicalrdquo layer iesensing communication data processing and application Sucha perspective allows the reader to obtain insights on specificchallenges at each layer as well as the interplay between thelayers Previous works instead typically focus on single aspectssuch as user recruitment data collection or methodologies tofoster and promote privacy This work does characterize suchaspects by specifically showing the role of each with respectto each ldquotechnologicalrdquo layer

B Timeline

This section presents the temporal evolution of the milestonesworks that contributed to shaping the crowdsensing paradigmFig 3 shows graphically the time of appearance and groupsthe research works into four categories

bull Mobile phone sensing crowdsourcing and anticipatorycomputing

bull Seminal worksbull Simulators and platforms

bull Privacy and trustTo uncover the time-evolution the following paragraphs

describe these fundamental works by year

2006 The concept of crowdsourcing appeared originally inan article written by Howe [57] that describes the rise ofthis paradigm giving an exact definition and presenting someof the first applications in this field Burke et al introducedparticipatory sensing as a novel paradigm in which people usetheir mobile devices to gather and share local knowledge [58]

2007 The use of mobile devices for automatic multimedia col-lection for specific application appeared with DietSense whereimage browsing processing and clustering are exploited tomanage health care and nutrition in participatory sensing [19]

2008 Miluzzo et al introduced the CenceMe applicationwhich is among the first systems that combine data collectionfrom sensors embedded in mobile devices with sharing onsocial networks [59] Micro-blog is one of the first applicationswhich aims at sharing and querying content through mobilephones and social participation Users are encouraged to recordmultimedia blogs that may be enriched with a variety of sensordata [60]

2010 Lane et al presented a survey about mobile phonesensing which includes a detailed description of sensors andexisting applications Furthermore the article shows how toscale from personal sensing to community sensing throughdata aggregation [8]

2011 Ganti et al proposed one among the first surveysthat are specific on MCS They uncover the potential of thecrowd where individuals with sensing and computing devicescollectively share data to measure and map phenomena ofcommon interest [7] Christin et al presented one among thefirst works analyzing the state-of-art in privacy-related concernsof participatory sensing systems [16]

2013 Cardone et al investigate how to foster the participationof citizens through a geo-social crowdsensing platform [61]The article focuses on three main technical aspects namelya geo-social model to profile users a matching algorithmfor participant selection and an Android-based platform toacquires data Khan et al surveyed existing works on mobilephone sensing and proposed one of the first classifications on




2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]

Phone sensing and crowdsourcing Seminal worksSimulators and Platforms Privacy and Trust

Fig 3 MCS timeline It presents the main works that have contributed to the evolution of MCS paradigm divided between mobile phone sensing andcrowdsourcing seminal works simulators and platforms privacy and trust

user involvement [9] Vastardis et al presented the existingarchitectures in mobile social networks their social propertiesand key research challenges [62] MOSDEN was one among thefirst collaborative sensing frameworks operating with mobilephones to capture and share sensed data between multipledistributed applications and users [63]

2014 A team in the University of Bologna launches theParticipAct Living Lab The core lab activities lead to oneof the first large-scale real-world experiments involving datacollection from sensors of smartphones of 200 students for oneyear in the Emilia Romagna region of Italy [64] Kantarciet al propose a reputation-based scheme to ensure datatrustworthiness where IoT devices can enhance public safetymanaging crowdsensing services provided by mobile deviceswith embedded sensors [65] The human factor as one of themost important components of MCS Ma et al investigatehow human mobility correlates with sensing coverage anddata transmission requirements [37] Guo et al define MobileCrowdsensing in [10] by specifying the required features for asystem to be defined as a MCS system The article providesa retrospective study of MCS paradigm as an evolution ofparticipatory sensing Pournajaf et al described the threats tousersrsquo privacy when personal information is disclosed andoutlined how privacy mechanisms are utilized in existingsensing applications to protect the participants [15] Tanas et alwere among the first to use ns-3 network simulator to assessthe performance of crowdsensing networks The work exploitsspecific features of ns-3 such as the mobility properties ofnetwork nodes combined with ad-hoc wireless interfaces [66]

2015 Guo et al investigate the interaction between machineand human intelligence in crowdsensing processes highlightingthe fundamental role of having humans in the loop [35]

2016 Chessa et al describe how to utilize socio-technicalnetworking aspects to improve the performance of MCS

MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS

Fig 4 Factors contributing to the rise of MCS

campaigns [67]

2017 Fiandrino et al introduce CrowdSenSim the firstsimulator for crowdsensing activities It features independentmodules to develop use mobility location communicationtechnologies and sensors involved according to the specificsensing campaign [68]

C Factors Contributing to the Rise of MCS

Several factors have contributed to the rise of MCS as oneof the most promising data collection paradigm (see Fig 4)

Mobile Phone Sensing As mentioned above mobile phonesensing is the forefather of MCS Unlike MCS the objectivephone sensing applications are at the individual level For




example mobile applications used for fitness utilize the GPSand other sensors (eg cardiometer) to monitor sessions [69]Ubifit encourages users to monitor their daily physical activitiessuch as running walking or cycling [70] Health care is anotherdomain in which individual monitoring is fundamental fordiet [19] or detection and reaction to elderly people falls [71]ndash[74] Other examples consist of distinguishing transportationmodes [75]ndash[78] recognition of speech [79] and drivingstyle [80] and for indoor navigation [81]ndash[83]Mobile smart devicesWearables The transition from mobilephones to smart mobile devices has boosted the rise of MCSWhile mobile phones only allowed phone calls and textmessages smartphones are equipped with sensing computingand communication capabilities In addition to smartphonestablets and wearables are other smart devices with the sameproperties With wearables user experience is fully includedin the technology process In [84] the authors present anapplication involving a WSN in a scenario focusing on sportand physical activities Wearables and body sensor networksare fundamental in health-care monitoring including sportsactivities medical treatments and nutrition [85] [86] In [87]the authors focus on inertial measurements units (IMUs) andwearable motion tracking systems for cost-effective motiontracking with a high impact in human-robot interactionCrowdsourcing Crowdsourcing is the key that has steeredphone sensing towards crowdsensing by enforcing community-oriented application purposes and leveraging large user par-ticipation for data collection According to the disciplineseveral definitions of crowdsourcing have been proposed Ananalysis of this can be found in [88]ndash[90] Howe in [57] coinedthe original term According to his definition crowdsourcingrepresents the act of a company or an institution in outsourcingtasks formerly performed by employees to an undefinednetwork of people in the form of an open call This enforcespeer-production when the job is accomplished collaborativelybut crowdsourcing is not necessarily only collaborative Amultitude of single individuals accepting the open call stillmatches with the definition Incentives mechanisms have beenproposed since the early days of crowdsourcing Several studiesconsider Amazonrsquos Mechanical Turk as one of the mostimportant examples of incentive mechanisms for micro-tasksthe paradigm allowing to engage a multitude of users for shorttime and low monetary cost [91] [92]Human factor The fusion between complementary roles ofhuman and machine intelligence builds on including humansin the loop of sensing computing and communicating pro-cesses [35] The impact of the human factor is fundamental inMCS for several reasons First of all mobility and intelligenceof human participants guarantee higher coverage and bettercontext awareness if compared to traditional sensor networksAlso users maintain by themselves the mobile devices andprovide periodic recharge Designing efficient human-in-the-loop architecture is challenging In [93] the authors analyzehow to exploit the human factor for example by steeringbehaviors after a learning phase Learning and predictiontypically require probabilistic methods [94]Cloud computing Considering the limited resources of localdata storage in smart devices processing such data usually

takes place in the cloud [95] [96] MCS is one of themost prominent applications in cloud-centric IoT systemswhere smart devices offer resources (the embedded sensingcapabilities) through cloud platforms on a pay-as-you-gobasis [97]ndash[99] Mobile devices contribute to a considerableamount of gathered information that needs to be stored foranalysis and processing The cloud enables to easily accessshared resources and common infrastructure with a ubiquitousapproach for efficient data management [100] [101]Internet of Things (IoT) The IoT paradigm is characterizedby a high heterogeneity of end systems devices and link layertechnologies Nonetheless MCS focuses on urban applicationswhich narrows down the scope of applicability of IoT Tosupport the smart cities vision urban IoT systems shouldimprove the quality of citizensrsquo life and provide added valueto the community by exploiting the most advanced ICTsystems [29] In this context MCS complements existingsensing infrastructures by including humans in the loopWireless Sensor Networks (WSN) Wireless sensor networksare sensing infrastructures employed to monitor phenomenaIn MCS the sensing nodes are human mobile devices AsWSN nodes have become more powerful with time multipleapplications can run over the same WSN infrastructure troughvirtualization [102] WSNs face several scalability challengesThe Software-Defined Networking (SDN) approach can beemployed to address these challenges and augment efficiencyand sustainability under the umbrella of software-definedwireless sensor networks (SDWSN) [103] The proliferation ofsmall sensors has led to the production of massive amounts ofdata in smart communities which cannot be effectively gatheredand processed in WSNs due to their weak communicationcapability Merging the concept of WSNs and cloud computingis a promising solution investigated in [104] In this work theauthors introduce the concept of sensors clouds and classify thestate of the art of WSNs by proposing a taxonomy on differentaspects such as communication technologies and type of data

D Acronyms used in the MCS Domain

MCS encompasses different fields of research and it existsdifferent ways of offering MCS services (see Table II)This Subsection overviews and illustrates the most importantmethods

S2aaS MCS follows a Sensing as a Service (S2aaS) businessmodel which makes data collected from sensors available tocloud users [98] [105] [111] Consequently companies andorganizations have no longer the need to acquire infrastructureto perform a sensing campaign but they can exploit existingones on a pay-as-you-go basis The efficiency of S2aaS modelsis defined in terms of the revenues obtained and the costs Theorganizer of a sensing campaign such as a government agencyan academic institution or a business corporation sustains coststo recruit and compensate users for their involvement [112] Theusers sustain costs while contributing data too ie the energyspent from the batteries for sensing and reporting data andeventually the data subscription plan if cellular connectivity isused for reporting




TABLE IIACRONYMS RELATED TO THE MOBILE CROWDSENSING DOMAIN

ACRONYM MEANING DESCRIPTION REFERENCES

S2aaS Sensing as a Service Provision to the public of data collected from sensors [98] [105]SenaaS Sensor as a Service Exposure of IoT cloudrsquos sensors capabilities and data in the form of services [106]SAaaS Sensing and Actuation as a Service Provision of simultaneous sensing and actuation resources on demand [107]MCSaaS Mobile CrowdSensing as a Service Extension and adaptation of the SAaaS paradigm to deploy and enable rapidly

mobile applications providing sensing and processing capabilities[108]

MaaS Mobility as a Service Combination of options from different transport providers into a single mobileservice

[109]

SIaaS Sensing Instrument as a Service Provision virtualized sensing instruments capabilities and shares them as acommon resource

[99]

MAaaS Mobile Application as a Service Provision of resources for storage processing and delivery of sensed data tocloud applications

[95]

MSaaS Mobile Sensing as a Service Sharing data collected from mobile devices to other users in the context of ITS [110]

SenaaS Sensor-as-a-Service has proposed to encapsulate bothphysical and virtual sensors into services according to ServiceOriented Architecture (SOA) [106] SenaaS mainly focuseson providing sensor management as a service rather thanproviding sensor data as a service It consists of three layersreal-world access layer semantic overlay layer and servicesvirtualization layer the real-world access layer is responsiblefor communicating with sensor hardware Semantic overlaylayer adds a semantic annotation to the sensor configurationprocess Services virtualization layer facilitates the users

SAaaS In SAaaS sensors and actuators guarantee tieredservices through devices and sensor networks or other sensingplatforms [107] [113] The sensing infrastructure is charac-terized by virtual nodes which are mobile devices owned bycitizens that join voluntarily and leave unpredictably accordingto their needs In this context clients are not passive interfacesto Cloud services anymore but they contribute through theircommunication sensing and computational capabilities

MCSaaS Virtualizing and customizing sensing resourcesstarting from the capabilities provided by the SAaaS modelallows for concurrent exploitation of pools of devices by severalplatformapplication providers [108] Delivering MCSaaS mayfurther simplify the provisioning of sensing and processingactivities within a device or across a pool of devices More-over decoupling the MCS application from the infrastructurepromotes the roles of a sensing infrastructure provider thatenrolls and manages contributing nodes

MaaS Mobility as a Service fuses different types of travelmodalities eg combines options from different transportproviders into a single mobile service to make easier planningand payments MaaS is an alternative to own a personal vehicleand makes it possible to exploit the best option for each journeyThe multi-modal approach is at the core of MaaS A travelercan exploit a combination of public transport bike sharingcar service or taxi for a single trip Furthermore MaaS caninclude value-added services like meal-delivery

SIaaS Sensing Instrument as a Service indicates the ideato exploit data collection instruments shared through cloudinfrastructure [99] It focuses on a common interface whichoffers the possibility to manage physical sensing instrument

while exploiting all advantages of using cloud computingtechnology in storing and processing sensed data

MAaaS Mobile Application as a Service indicates a model todeliver data in which the cloud computing paradigm providesresources to store and process sensed data specifically focusedon applications developed for mobile devices [114]

MSaaS Mobile Sensing as a Service is a concept based onthe idea that owners of mobile devices can join data collectionactivities and decide to share the sensing capabilities of theirmobile devices as a service to other users [110] Specificallythis approach refers to the interaction with Intelligent Trans-portation Systems (ITS) and relies on continuous sensing fromconnected cars The MSaaS way of delivering MCS servicecan constitute an efficient and flexible solution to the problemof real-time traffic management and data collection on roadsand related fields (eg road bumping weather condition)

III MOBILE CROWDSENSING IN A NUTSHELL

This Section presents the main characteristic of MCS Firstwe introduce MCS as a layered architecture discussing eachlayer Then we divide MCS data collection frameworks be-tween domain-specific and general-purpose and survey the mainworks After that theoretical works on operational researchand optimization problems are presented and classified Thenwe present platforms simulators and dataset Subsec III-Fcloses the section by providing final remarks and outlines thenew taxonomies that we propose for each layer These are themain focus of the upcoming Sections

A MCS as a Layered Architecture

Similarly to [40] [42] we present a four-layered architectureto describe the MCS landscape as shown in Fig 1 Unlikethe rationale in [40] [42] our proposal follows the directionof command and control From top to bottom the first layeris the Application layer which involves user- and task-relatedaspects The second layer is the Data layer which concernsstorage analytics and processing operations of the collectedinformation The third layer is the Communication layerthat characterizes communication technologies and reportingmethodologies Finally at the bottom of the layered architecturethere is the Sensing layer that comprises all the aspects




BP

MCS campaignorganizer

User recruitmentamp selection

Task allocationamp distribution

Fig 5 Application layer It comprises high-level task- and user-related aspectsto design and organize a MCS campaign

Fog Fog

Raw Data

Cloud

Data Analysis Processing and Inference

Fig 6 Data layer It includes technologies and methodologies to manage andprocess data collected from users

involving the sensing process and modalities In the followingSections this architecture will be used to propose differenttaxonomies that will considered several aspects for each layerand according to classification to overview many MCS systems(Sec IV Sec V Sec VI and Sec VII)

Application layer Fig 5 shows the application layer whichinvolves high-level aspects of MCS campaigns Specifically thephases of campaign design and organization such as strategiesfor recruiting users and scheduling tasks and approachesrelated to task accomplishment for instance selection ofuser contributions to maximize the quality of informationwhile minimizing the number of contributions Sec IV willpresent the taxonomies on the application layer and the relatedclassifications

Data layer Fig 6 shows the data layer which comprises allcomponents responsible to store analyze and process datareceived from contributors It takes place in the cloud orcan also be located closer to end users through fog serversaccording to the need of the organizer of the campaign For

Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station

WiFi Access PointIndividualReporting

CollaborativeReportingCellular

CollaborativeReporting

WiFi

Fig 7 Communication layer It comprises communication technologies anddata reporting typologies to deliver acquired data to the central collector

Embedded Sensors Connected Sensors

Smart Devices

Fig 8 Sensing layer It includes the sensing elements needed to acquire datafrom mobile devices and sampling strategies to efficiently gather it

instance in this layer information is inferred from raw data andthe collector computes the utility in receiving a certain typeof data or the quality of acquired information Taxonomies ondata layer and related overview on works will be presented inSec V

Communication layer Fig 7 shows the communication layerwhich includes both technologies and methodologies to deliverdata acquired from mobile devices through their sensors tothe cloud collector The mobile devices are typically equippedwith several radio interfaces eg cellular WiFi Bluetoothand several optimizations are possible to better exploit thecommunication interfaces eg avoid transmission of duplicatesensing readings or coding redundant data Sec VI will proposetaxonomies on this layer and classify works accordingly

Sensing layer Fig 8 shows the sensing layer which is atthe heart of MCS as it describes sensors Mobile devicesusually acquire data through built-in sensors but for specializedsensing campaigns other types of sensors can be connectedSensors embedded in the device are fundamental for its




normal utilization (eg accelerometer to automatically turnthe orientation of the display or light sensor to regulate thebrightness of the monitor) but can be exploited also to acquiredata They can include popular and widespread sensors such asgyroscope GPS camera microphone temperature pressurebut also the latest generation sensors such as NFC Specializedsensors can also be connected to mobile devices via cableor wireless communication technologies (eg WiFi direct orBluetooth) such as radiation gluten and air quality sensorsData acquired through sensors is transmitted to the MCSplatform exploiting the communication capabilities of mobiledevices as explained in the following communication layerSec VII will illustrate taxonomies proposed for this layer andconsequent classification of papers

B Data Collection Frameworks

The successful accomplishment of a MCS campaign requiresto define with precision a number of steps These dependon the specific MCS campaign purpose and range from thedata acquisition process (eg deciding which sensors areneeded) to the data delivery to the cloud collector (eg whichcommunication technologies should be employed) The entireprocess is known as data collection framework (DCF)

DCFs can be classified according to their scope in domain-specific and general-purpose A domain-specific DCF is specif-ically designed to monitor certain phenomena and providessolutions for a specific application Typical examples areair quality [21] noise pollution monitoring [115] and alsohealth care such as the gluten sensor [116] for celiacsViceversa General-purpose DCFs are not developed for aspecific application but aim at supporting many applications atthe same time

Domain-specific DCFs Domain-specific DCFs are specificallydesigned to operate for a given application such as healthcare environmental monitoring and intelligent transportationamong the others One of the most challenging issues insmart cities is to create a seamless interconnection of sensingdevices communication capabilities and data processingresources to provide efficient and reliable services in manydifferent application domains [170] Different domains ofapplicability typically correspond to specific properties anddesign requirements for the campaign eg exploited sensorsor type of data delivery For instance a noise monitoringapplication will use microphone and GPS and usually isdelay-tolerant while an emergency response application willtypically require more sensors and real-time data reporting Inother words domain-specific DCFs can be seen as use casesof MCS systems The following Subsection presents the mostpromising application domains where MCS can operate toimprove the citizensrsquo quality of life in a smart city domainand overview existing works as shown in Table IIIEmergency management and prevention This category com-prises all application related to monitoring and emergencyresponse in case of accidents and natural disasters such asflooding [24] earthquakes [117]ndash[119] fires and nucleardisasters [120] For instance Creekwatch is an applicationdeveloped by the IBM Almaden research center [24] It allows

to monitor the watershed conditions through crowdsensed datathat consists of the estimation of the amount of water in theriver bed the amount of trash in the river bank the flow rateand a picture of the waterway

Environmental Monitoring Environmental sensing is fundamen-tal for sustainable development of urban spaces and to improvecitizensrsquo quality of life It aims to monitor the use of resourcesthe current status of infrastructures and urban phenomena thatare well-known problems affecting the everyday life of citizenseg air pollution is responsible for a variety of respiratorydiseases and can be a cause of cancer if individuals are exposedfor long periods To illustrate with some examples the PersonalEnvironment Impact Report (PEIR) exploits location datasampled from everyday mobile phones to analyze on a per-userbasis how transportation choices simultaneously impact on theenvironment and calculate the risk-exposure and impact forthe individual [121] Ear-Phone is an end-to-end urban noisemapping system that leverages compressive sensing to solvethe problem of recovering the noise map from incompleteand random samples [122] This platform is delay tolerantand exploits only WiFi because it aims at creating maps anddoes not need urgent data reporting NoiseTube is a noisemonitoring platform that exploits GPS and microphone tomeasure the personal exposure of citizen to noise in everydaylife [123] NoiseMap measures value in dB corresponding tothe level of noise in a given location detected via GPS [115]In indoor environments users can tag a particular type ofnoise and label a location to create maps of noise pollutionin cities by exploiting WiFi localization [124] In [125] theauthors propose a platform that achieves tasks of environmentalmonitoring for smart cities with a fine granularity focusingon data heterogeneity Nowadays air pollution is a significantissue and exploiting mobility of humans to monitor air qualitywith sensors connected to their devices may represent a win-win solution with higher accuracy than fixed sensor networksHazeWatch is an application developed in Sydney that relieson active citizen participation to monitor air pollution andis currently employed by the National Environment Agencyof Singapore on a daily basis [22] GasMobile is a smalland portable measurement system based on off-the-shelfcomponents which aims to create air pollution maps [21]CommonSense allows citizens to measure their exposure at theair pollution and in forming groups to aggregate the individualmeasurements [56] Counter-Strike consists of an iterativealgorithm for identifying low-level radiation sources in theurban environment which is hugely important for the securityprotection of modern cities [126] Their work aim at makingrobust and reliable the possible inaccurate contribution of usersIn [127] the authors exploit the flash and the camera of thesmartphone as a light source and receptor In collaboration witha dedicated sensor they aim at measuring the amount of dust inthe air Recent discoveries of signature in the ionosphere relatedto earthquakes and tsunamis suggests that ionosphere may beused as a sensor that reveals Earth and space phenomena [128]The Mahali Project utilizes GPS signals that penetrate theionosphere to enable a tomographic analysis of the ionosphereexploited as a global earth system sensor [129]




TABLE IIIDOMAIN-SPECIFIC DATA COLLECTION FRAMEWORKS (DCFS)

DOMAIN OF INTEREST DESCRIPTION REFERENCES

Emergency prevention and management Prevention of emergencies (eg monitoring the amount of water in the river bed)and post-disaster management (earthquakes or flooding)

[24] [117]ndash[120]

Environmental monitoring Monitoring of resources and environmental conditions such as air and noisepollution radiation

[21] [22] [56] [115] [121]ndash[129]

E-commerce Collection sharing and live-comparison of prices of goods from real stores orspecific places such as gas stations

[130]ndash[133]

Health care amp wellbeing Sharing of usersrsquo physical or mental conditions for remote feedback or exchangeof information about wellbeing like diets and fitness

[17] [19] [116] [134]ndash[136]

Indoor localization Enabling indoor localization and navigation by means of MCS systems in GPS-denied environments

[137]ndash[139]

Intelligent transportation systems Monitoring of citizen mobility public transport and services in cities eg trafficand road condition available parking spots bus arrival time

[20] [140]ndash[145]

Mobile social networks Establishment of social relations meeting sharing experiences and data (photo andvideo) of users with similar interests

[62] [146]ndash[155]

Public safety Citizens can check share and evaluate the level of crimes for each areas in urbanenvironments

[156] [157]

Unmanned vehicles Interaction between mobile users and driver-less vehicles (eg aerial vehicles orcars) which require high-precision sensors

[158]ndash[160]

Urban planning Improving experience-based decisions on urbanization issues such as street networksdesign and infrastructure maintenance

[30] [161] [162]

Waste management Citizens help to monitor and support waste-recycling operations eg checking theamount of trash or informing on dynamic waste collection routing

[25] [26]

WiFi characterization Mapping of WiFi coverage with different MCS techniques such as exploitingpassive interference power measuring spectrum and received power intensity

[163]ndash[165]

Others Specific domain of interest not included in the previous list such as recommendingtravel packages detecting activity from sound patterns

[147] [148] [166]ndash[169]

E-commerce Websites comparing prices of goods and servicesguide customer decisions and improve competitiveness butthey are typically limited to the online world where most of theinformation is accessible over the Internet MCS can fill the gapconnecting physical and digital worlds and allowing customersto report information and compare pricing in different shops orfrom different vendors [130] To give a few examples cameraand GPS used in combination may track and compare fuelprices between different gas stations [131] While approachinggas stations an algorithm extracts the price from picturesrecorded through the camera and associates it with the gasstation exploiting the GPS Collected information is comparedwith the one gathered by other users to detect most convenientpetrol stations In [130] the authors present a participatorysensing paradigm which can be employed to track pricedispersion of similar consumer goods even in offline marketsMobishop is a distributed computing system designed tocollect process and deliver product price information fromstreet-side shops to potential buyers on their mobile phonesthrough receipt scanning [132] LiveCompare represents anotherexample of MCS solution that uses the camera to take picturesof bar codes and GPS for the market location to compare pricesof goods in real-time [133]

Health Care and wellbeing Health care allows diagnosingtreating and preventing illness diseases and injuries Itincludes a broad umbrella of fields connected to health thatranges from self-diagnostics to physical activities passingthrough diet monitoring HealthAware is a system that uses theaccelerometer to monitor daily physical activity and camerato take pictures of food that can be shared [17] SPA is asmartphone assisted chronic illness self-management systemthat facilitates patient involvement exploiting regular feedbacksof relevant health data [134] Yi et al propose an Android-

based system that collects displays and sends acquired datato the central collector [135] Dietsense automatically capturespictures of food GPS location timestamp references and amicrophone detects in which environment the sample wascollected for dietary choices [19] It aims to share data tothe crowd and allows just-in-time annotation which consistsof reviews captured by other participants AndWellness is apersonal data collection system that exploits the smartphonersquossensors to collect and analyze data from user experiences [136]

Indoor localization Indoor localization refers to the processof providing localization and navigation services to users inindoor environments In such a scenario the GPS does notwork and other sensors are employed Fingerprinting is apopular technique for localization and operates by constructinga location-dependent fingerprint map MobiBee collects finger-prints by exploiting quick response (QR) codes eg postedon walls or pillars as location tags [137] A location markeris designed to guarantee that the fingerprints are collectedat the targeted locations WiFi fingerprinting presents somedrawbacks such as a very labor-intensive and time-consumingradio map construction and the Received Signal Strength (RSS)variance problem which is caused by the heterogeneity ofdevices and environmental context instability In particular RSSvariance severely degrades the localization accuracy RCILS isa robust crowdsourcing indoor localization system which aimsat constructing the radio map with smartphonersquos sensors [138]RCILS abstracts the indoor map as a semantics graph in whichthe edges are the possible user paths and the vertexes are thelocation where users perform special activities Fraud attacksfrequently compromise reference tags employed for gettinglocation annotations Three types of location-related attacksin indoor MCS are considered in [139] including tag forgerytag misplacement and tag removal To deal with these attacks




the authors propose location-dependent fingerprints as supple-mentary information for improving location identification atruth discovery algorithm to detect falsified data and visitingpatterns to reveal tag misplacement and removal

Intelligent Transportation Systems MCS can also supportITS to provide innovative services improve cost-effectivenessand efficiency of transportation and traffic management sys-tems [140] It includes all the applications related to trafficvehicles public transports and road conditions In [141]social networks are exploited to acquire direct feedbackand potentially valuable information from people to acquireawareness in ITS In particular it verifies the reliability ofpollution-related social networks feedbacks into ITS systemsIn [142] the authors present a system that predicts bus arrivaltimes and relies on bus passengersrsquo participatory sensing Theproposed model is based only on the collaborative effort of theparticipants and it uses cell tower signals to locate the userspreventing them from battery consumption Furthermore ituses accelerometer and microphone to detect when a user is ona bus Wind warning systems alert drivers while approachingbridges in case of dangerous wind conditions WreckWatchis a formal model that automatically detects traffic accidentsusing the accelerometer and acoustic data immediately send anotification to a central emergency dispatch server and providephotographs GPS coordinates and data recordings of thesituation [143] Nericell is a system used to monitor roadand traffic conditions which uses different sensors for richsensing and detects potholes bumps braking and honking [20]It exploits the piggybacking mechanisms on usersrsquo smartphonesto conduct experiments on the roads of Bangalore Safestreetdetects and reports the potholes and surface abnormalities ofroads exploiting patterns from accelerometer and GPS [144]The evaluation exploits datasets collected from thousands ofkilometers of taxi drives in Mumbai and Boston VTrack is asystem which estimates travel time challenging with energyconsumption and inaccurate position samples [145] It exploitsa HMM (Hidden Markov Model)-based map matching schemeand travel time estimation method that interpolates sparse datato identify the most probable road segments driven by the userand to attribute travel time to those segments

Mobile Social Networks (MSNs) MSNs are communicationsystems that allow people with similar interests to establishsocial relations meet converse and share exploiting mobiledevices [62] [146] In this category the most fundamentalaspect is the collaboration between users who share thedata sensed through smartphone sensors such as recognizedactivities and locations Crowdsenseplace is a system thatfocuses on scaling properties of place-centric crowdsensingto provide place related informations [147] including therelationship between users and coverage [148] MSNs focusnot only on the behavior but also on the social needs ofthe users [62] The CenceMe application is a system whichcombines the possibility to use mobile phone embedded sensorsfor the sharing of sensed and personal information throughsocial networking applications [59] It takes a user status interms of his activity context or habits and shares them insocial networks Micro-Blog is a location-based system to

automatically propose and compare the context of the usersby leveraging new kinds of application-driven challenges [60]EmotionSense is a platform for social psychological studiesbased on smartphones [149] its key idea is to map notonly activities but also emotions and to understand thecorrelation between them It gathers user emotions as wellas proximity and patterns of conversations by processing theaudio from the microphone MobiClique is a system whichleverages already existing social networks and opportunisticcontacts between mobile devices to create ad hoc communitiesfor social networking and social graph based opportunisticcommunications [150] MIT Serendipity is one of the firstprojects that explored the MSN aspects [151] it automatesinformal interactions using Bluetooth SociableSense is aplatform that realizes an adaptive sampling scheme basedon learning methods and a dynamic computation distributionmechanism based on decision theory [152] This systemcaptures user behavior in office environments and providesthe participants with a quantitative measure of sociability ofthem and their colleagues WhozThat is a system built onopportunistic connectivity for offering an entire ecosystem onwhich increasingly complex context-aware applications canbe built [153] MoVi a Mobile phone-based Video highlightssystem is a collaborative information distillation tool capableof filtering events of social relevance It consists of a triggerdetection module that analyses the sensed data of several socialgroups and recognizes potentially interesting events [154]Darwin phones is a work that combines collaborative sensingand machine learning techniques to correlate human behaviorand context on smartphones [155] Darwin is a collaborativeframework based on classifier evolution model pooling andcollaborative sensing which aims to infer particular momentsof peoplersquos life

Public Safety Nowadays public safety is one of the most criticaland challenging issues for the society which includes protectionand prevention from damages injuries or generic dangersincluding burglary trespassing harassment and inappropriatesocial behaviors iSafe is a system for evaluating the safetyof citizens exploiting the correlation between informationobtained from geo-social networks with crime and census datafrom Miami county [156] In [157] the authors investigatehow public safety could leverage better commercial IoTcapabilities to deliver higher survivability to the warfighter orfirst responders while reducing costs and increasing operationalefficiency and effectiveness This paper reviews the maintactical requirements and the architecture examining gaps andshortcomings in existing IoT systems across the military fieldand mission-critical scenarios

Unmanned Vehicles Recently unmanned vehicles have becomepopular and MCS can play an important role in this fieldIndeed both unmanned aerial vehicles (UAVs) and driver-lesscars are equipped with different types of high-precision sensorsand frequently interact with mobile devices and users [158]In [159] the authors investigate the problem of energy-efficientjoint route planning and task allocation for fixed-wing UAV-aided MCS system The corresponding joint optimizationproblem is developed as a two-stage matching problem In




the first stage genetic algorithms are employed to solve theroute planning In the second stage the proposed Gale-Shapleyalgorithm solves the task assignment To provide a global viewof traffic conditions in urban environments [160] proposesa trust-aware MCS technique based on UAVs The systemreceives real traffic information as input and UAVs build thedistribution of vehicles and RSUs in the network

Urban planning Urban planning in the context of MCS isrelated to improve experience-based decisions on urbanizationissues by analyzing data acquired from different devices ofcitizens in urban environments It aims to improve citizensrsquoquality of life by exploiting sensing technologies data an-alytics and processing tools to deploy infrastructures andsolutions [161] For instance some studies investigate theimpact of the street networks on the spatial distribution ofurban crimes predicting that violence happens more often onwell-connected roads [162] In [30] the authors show that dataacquired from accelerometers embedded in smartphones ofcar drivers can be used for monitoring bridge vibrations bydetecting several modal frequenciesWaste Management Although apparently related to environmen-tal monitoring we classify waste management in a standalonecategory on purpose Waste management is an emerging fieldwhich aims to handle effectively waste collection ie how toappropriately choose location of bins and optimal routes ofcollecting trucks recycling and all the related processes [171]WasteApp presents a design and an implementation of a mobileapp [26] It aims at merging behavioral studies and standardfeatures of existing mobile apps with a co-design methodologythat gathers real user needs for waste recycling Garbage Watchemploys citizens to monitor the content of recycling bins toenhance recycling program [25]WiFi characterization This application characterizes the WiFicoverage in a certain area for example by measuring spectrumand interference [164] In [163] the authors propose a modelfor sensing the volume of wireless activity at any frequencyexploiting the passive interference power This techniqueutilizes a non-intrusive way of inferring the level of wirelesstraffic without extracting data from devices Furthermorethe presented approach is independent of the traffic patternand requires only approximate location information MCNetenables WiFi performance measurements and mapping dataabout WLAN taken from users that participate in the sensingprocess [165]Others This category includes works not classified in theprevious fields of application but still focusing on specificdomains SoundSense is a scalable framework for modelingsound events on smartphones using a combination of super-vised and unsupervised learning techniques to classify bothgeneral sound types and discover sound events specific toindividual users [166] ConferenceSense acquires data extractand understand community properties to sense large eventslike conferences [167] It uses some sensors and user inputsto infer contexts such as the beginning and the end of agroup activity In [169] the authors propose travel packagesexploiting a recommendation system to help users in planningtravels by leveraging data collected from crowdsensing They

distinguish user preferences extract points of interest (POI)and determine location correlations from data collected Thempersonalized travel packages are determined by consideringpersonal preferences POI temporal and spatial correlationsImprove the Location Reliability (ILR) is a scheme in whichparticipatory sensing is used to achieve data reliability [168]The key idea of this system is to validate locations usingphoto tasks and expanding the trust to nearby data points usingperiodic Bluetooth scanning The participants send photo tasksfrom the known location which are manually or automaticallyvalidated

General-purpose DCFs General-purpose DCFs investigatecommon issues in MCS systems independently from theirdomains of interest For instance energy efficiency and task cov-erage are two fundamental factors to investigate independentlyif the purpose of a campaign is environmental monitoringhealth care or public safety This Subsection presents the mainaspects of general-purpose DCFs and classifies existing worksin the domain (see Table IV for more insights)Context awareness Understanding mobile device context isfundamental for providing higher data accuracy and does notwaste battery of smartphones when sensing does not meet theapplication requirements (eg taking pictures when a mobiledevice is in a pocket) Here-n-now is a work that investigatesthe context-awareness combining data mining and mobileactivity recognition [172] Lifemap provides location-basedservices exploiting inertial sensors to provide also indoorlocation information [173] Gathered data combines GPS andWiFi positioning systems to detect context awareness better Toachieve a wide acceptance of task execution it is fundamentala deep understanding of factors influencing user interactionpatterns with mobile apps Indeed they can lead to moreacceptable applications that can report collected data at theright time In [194] the authors investigate the interactionbehavior with notifications concerning the activity of usersand their location Interestingly their results show that userwillingness to accept notifications is strongly dependent onlocation but only with minor relation to their activitiesEnergy efficiency To foster large participation of users it isnecessary that sensing and reporting operations do not draintheir batteries Consequently energy efficiency is one of themost important keys to the success of a campaign PiggybackCrowdsensing [174] aims to lower the energy overhead ofmobile devices exploiting the opportunities in which users placephone calls or utilize applications Devices do not need to wakeup from an idle state to specifically report data of the sensingcampaign saving energy Other DCFs exploit feedback fromthe collector to avoid useless sensing and reporting operationsIn [175] a deterministic distributed DCF aims to foster energy-efficient data acquisition in cloud-based MCS systems Theobjective is to maximize the utility of the central collector inacquiring data in a specific area of interest from a set of sensorswhile minimizing the battery costs users sustain to contributeinformation A distributed probabilistic algorithm is presentedin [176] where a probabilistic design regulates the amount ofdata contributed from users in a certain region of interest tominimize data redundancy and energy waste The algorithm is




TABLE IVGENERAL-PURPOSE DATA COLLECTION FRAMEWORKS (DCFS)

TARGET DESCRIPTION REFERENCES

Context awareness Combination of data mining and activity recognition techniques for contextdetection

[172] [173]

Energy efficiency Strategies to lower the battery drain of mobile devices during data sensing andreporting

[174]ndash[178]

Resource allocation Strategies for efficient resource allocation during data contribution such aschannel condition power spectrum computational capabilities

[179]ndash[181]

Scalability Solutions to develop DCFs with good scalability properties during run-time dataacquisition and processing

[182] [183]

Sensing task coverage Definition of requirements for task accomplishment such as spatial and temporalcoverage

[184]ndash[187]

Trustworthiness and privacy Strategies to address issues related to preserve privacy of the contributing usersand integrity of reported data

[188]ndash[193]

based on limited feedback from the central collector and doesnot require users to complete a specific task EEMC (Enablingenergy-efficient mobile crowdsensing) aims to reduce energyconsumption due to data contribution both for individuals andthe crowd while making secure the gathered information froma minimum number of users within a specific timeframe [177]The proposed framework includes a two-step task assignmentalgorithm that avoids redundant task assignment to reduce theaverage energy consumption In [178] the authors provide acomprehensive review of energy saving techniques in MCS andidentify future research opportunities Specifically they analyzethe main causes of energy consumption in MCS and presenta general energy saving framework named ESCrowd whichaims to describe the different detailed MCS energy savingtechniques

Resource allocation Meeting the demands of MCS cam-paigns requires to allocate resources efficiently The predictiveChannel-Aware Transmission (pCAT) scheme is based on thefact that much less spectrum resource allocation is requiredwhen the channel is in good condition [179] Considering usertrajectories and recurrent spots with a good channel conditionthe authors suggest that background communication trafficcan be scheduled by the client according to application datapriority and expected quality data In [180] the authors considerprecedence constraints in specialized tasks among multiplesteps which require flexibility in order to the varying level oftheir dependency Furthermore they consider variations of loadsneeded for accomplishing different tasks and require allocationschemes that are simple fast decentralized and provide thepossibility to choose contributing users The main contributionof authors is to focus on the performance limits of MCS systemsregarding these issues and propose efficient allocation schemesclaiming they meet the performance under the consideredconstraints SenSquare is a MCS architecture that aims tosave resources for both contributors and stakeholders reducingthe amount of data to deliver while maintaining its precisionand rewarding users for their contribution [181] The authorsdeveloped a mobile application as a case study to prove theeffectiveness of SenSquare by evaluating the precision ofmeasures and resources savings

Scalability MCS systems require large participation of usersto be effective and make them scalable to large urban envi-ronments with thousands of citizens is paramount In [182]the authors investigate the scalability of data collection andrun-time processing with an approach of mobileonboarddata stream mining This framework provides efficiency inenergy and bandwidth with no consistent loss of informationthat mobile data stream mining could obtain In [183] theauthors investigate signicant issues in MCS campaigns such asheterogeneity of mobile devices and propose an architecture tolower the barriers of scalability exploiting VM-based cloudlets

Sensing task coverage To accomplish a task a sensing cam-paign organizer requires users to effectively meet applicationdemands including space and time coverage Some DCFsaddress the problems related to spatio-temporal task coverageA solution to create high-accurate monitoring is to designan online scheduling approach that considers the location ofdevices and sensing capabilities to select contributing nodesdeveloping a multi-sensor platform [184] STREET is a DCFthat investigates the problem of the spatial coverage classifyingit between partial coverage full coverage or k-coverage It aimsto improve the task coverage in an energy-efficient way byrequiring updates only when needed [185] Another approachto investigate the coverage of a task is to estimate the optimalnumber of citizens needed to meet the sensing task demandsin a region of interest over a certain period of time [186]Sparse MCS aims to lower costs (eg user rewarding andenergy consumption) while ensuring data quality by reducingthe required number of sensing tasks allocated [187] To thisend it leverages the spatial and temporal correlation amongthe data sensed in different sub-areas

Trustworthiness and Privacy As discussed a large participationof users is essential to make effective a crowdsensing campaignTo this end one of the most important factors for loweringbarriers of citizens unwillingness in joining a campaign isto guarantee their privacy On the other hand an organizerneeds to trust participants and be sure that no maliciousnodes contribute unreliable data In other words MCS systemsrequire mechanisms to guarantee privacy and trustworthinessto all components There exist a difference between the terms




trustworthiness and truthfulness when it comes to MCS Theformer indicates how valuable and trusted is data contributedfrom users The latter indicates how truthful a user is whenjoining the auction because a user may want to increase thedata utility to manipulate the contribution

In MCS systems the problem of ensuring trustworthinessand privacy remains open First data may be often unreliabledue to low-quality sensors heterogeneous sources noise andother factors Second data may reveal sensitive informationsuch as pictures mobility patterns user behavior and individualhealth Deco is a DCF that investigates malicious and erroneouscontributions which inevitably result in poor data quality [189]It aims to detect and correct false and missing data by applyingspatio-temporal compressive sensing techniques In [188] theauthors propose a framework whose objective is twofoldFirst to extract reliable information from heterogeneous noisyand biased data collected from different devices Then topreserve usersrsquo privacy To reach these goals they design anon-interactive privacy-preserving truthful-discovery systemthat follows a two-server model and leverages Yaorsquos Garbledcircuit to address the problem optimization BUR is a BasicUser Recruitment protocol that aims to address privacy issuesby preserving user recruitment [190] It is based on a greedystrategy and applies secret sharing techniques to proposea secure user recruitment protocol In [191] the authorsinvestigate privacy concerns related to incentive mechanismsfocusing on Sybil attack where a user may illegitimately pretendmultiple identities to receive benefits They design a Sybil-proofincentive mechanism in online scenarios that depends on usersrsquoflexibility in performing tasks and present single-minded andmulti-minded cases DTRPP is a Dynamic Trust Relationshipsaware data Privacy Protection mechanism that combines keydistribution with trust management [192] It is based on adynamic management of nodes and estimation of the trustdegree of the public key which is provided by encounteringnodes QnQ is a Quality and Quantity based unified approachthat proposes an event-trust and user-reputation model toclassify users according to different classes such as honestselfish or malicious [193] Specifically QnQ is based on arating feedback scheme that evaluates the expected truthfulnessof specific events by exploiting a QoI metric to lower effectsof selfish and malicious behaviors

C Theoretical Works Operational Research and Optimization

This Subsection presents analytical and theoretical studiesproposed in the domain of MCS such as operational researchand optimization problems While the previous Subsectiondiscusses the technicalities of data collection this Subsectionpresents research works that analyze MCS systems from atheoretical perspective We classify these works according totheir target (see Table V)

We remark that this Subsection differentiates from theprevious ones because it focuses on aspects like abstractproblem formulation key challenges and proposed solutionsFor instance while the previous Section presents data collectionframeworks that acquire data leveraging context awarenessthis Section presents a paragraph on context awareness that

investigates how to formulate the problem and the proposedoptimized solutions without practical implementation

Trade-off between amount and quality data with energyconsumption In the last years many researchers have focusedtheir attention on the trade-off between the amount and qualityof collected data and the corresponding energy consumptionThe desired objectives are

bull to obtain high quality of contributed data (ie to maximizethe utility of sensing) with low energy consumption

bull to limit the collection of low-quality dataThe Minimum Energy Single-sensor task Scheduling (MESS)

problem [196] analyzes how to optimally schedule multiplesensing tasks to the same user by guaranteeing the qual-ity of sensed data while minimizing the associated energyconsumption The problem upgrades to be Minimum EnergyMulti-sensor task Scheduling (MEMS) when sensing tasksrequire the use of multiple sensors simultaneously In [195]the authors investigate task allocation and propose a first-in-first-out (FIFO) task model and an arbitrary deadline (AD) taskmodel for offline and online settings The latter scenario ismore complex because the requests arrive dynamically withoutprior information The problem of ensuring energy-efficiencywhile minimizing the maximum aggregate sensing time is NP-hard even when tasks are defined a priori [197] The authorsfirst investigate the offline allocation model and propose anefficient polynomial-time approximation algorithm with a factorof 2minus 1m where m is the number of mobile devices joiningthe system Then focusing on the online allocation modelthey design a greedy algorithm which achieves a ratio of atmost m In [36] the authors present a distributed algorithmfor maximizing the utility of sensing data collection when thesmartphone cost is constrained The design of the algorithmconsiders a stochastic network optimization technique anddistributed correlated scheduling

Sensing Coverage The space and temporal demands of a MCScampaign define how sensing tasks are generated Followingthe definition by He et al [241] the spatial coverage is definedas the total number of different areas covered with a givenaccuracy while the temporal coverage is the minimum amountof time required to cover all regions of the area

By being budget-constrained [198] investigates how tomaximize the sensing coverage in regions of interest andproposes an algorithm for sensing task allocation The articlealso shows considerations on budget feasibility in designingincentive mechanisms based on a scheme which determinesproportional share rule compensation To maximize the totalcollected data being budged-constrained in [203] organizerspay participants and schedule their sensing time based ontheir bids This problem is NP-hard and the authors propose apolynomial-time approximation

The effective target coverage measures the capability toprovide sensing coverage with a certain quality to monitor aset of targets in an area [199] The problem is tackled withqueuing model-based algorithm where the sensors arrive andleave the coverage region following a birth process and deathprocess The α T -coverage problem [201] consists in ensuringthat each point of interest in an area is sensed by at least one




TABLE VTHEORETICAL WORKS ON OPERATIONAL RESEARCH AND OPTIMIZATION PROBLEMS

TARGET OBJECTIVE REFERENCES

Trade-off data vs energy Maximization of the amount and quality of gathered data while minimizing theenergy consumption of devices

[36] [195]ndash[197]

Sensing coverage Focus on how to efficiently address the requirements on task sensing coveragein space and temporal domains

[198]ndash[205]

Task allocation Efficient task allocation among participants leveraging diverse techniques andapproaches

[206]ndash[217]

User recruitment Efficient user recruitment to meet the requirements of a sensing campaign whileminimizing the cost

[218]ndash[227]

Context awareness It consists in exploiting context-aware sensing to improve system performancein terms of delay bandwidth and energy efficiency

[228]ndash[235]

Budget-constrained Maximization of task accomplishment under budget constraints or minimizationof budget to fully accomplish a task

[112] [236]ndash[239] [240]

node with a minimum probability α during the time interval T The objective is to minimize the number of nodes to accomplisha task and two algorithms are proposed namely inter-locationand inter-meeting-time In the first case the algorithm selects aminimal group of nodes by considering the reciprocal distanceThe second one considers the expected meeting time betweenthe nodes

Tasks are typically expected to be fulfilled before a deadlineThis problem is considered in [202] where the participantsperform tasks with a given probability Cooperation amongthe participants to accomplish a common task ensures that thecompletion deadline is respected The deadline-sensitive userrecruitment problem is formalized as a set cover problem withnon-linear programming constraints which is NP-hard In [200]the authors focus on the allocation of location-dependent taskswhich is a NP-hard and propose a local-ratio-based algorithm(LRBA) The algorithm divides the whole problem into severalsub-problems through the modification of the reward function ateach iteration In [205] the authors aim to maximize the spatialdistribution and visual correlation while selecting geo-taggedphotos with the maximum utility The KL divergence-basedclustering can be employed to distinguish uncertain objectswith multiple features [204] The symmetry KL divergenceand Jensen-Shannon KL divergence are seen as two differentmutated forms to improve the proposed algorithm

Task allocation The problem of task allocation defines themethodologies of task dispatching and assignment to theparticipants TaskMe [209] solves two bi-objective optimizationproblems for multi-task user selection One considers thecase of a few participants with several tasks to accomplishThe second examines the case of several participants withfew tasks iCrowd is a unified task allocation framework forpiggyback crowdsensing [208] and defines a novel spatial-temporal coverage metric ie the k-depth coverage whichconsiders both the fraction of subareas covered by sensorreadings and the number of samples acquired in each coveredsub-area iCrowd can operate to maximize the overall k-depthcoverage across a sensing campaign with a fixed budget orto minimize the overall incentive payment while ensuring apredefined k-depth coverage constraint Xiao et al analyzethe task assignment problem following mobility models of

mobile social networks [210] The objective is to minimizethe average makespan Users while moving can re-distributetasks to other users via Bluetooth or WiFi that will deliverthe result during the next meeting Data redundancy canplay an important role in task allocation To maximize theaggregate data utility while performing heterogeneous sensingtasks being budget-constrained [211] proposes a fairness-awaredistributed algorithm that takes data redundancy into account torefine the relative importance of tasks over time The problemis subdivided it into two subproblems ie recruiting andallocation For the former a greedy algorithm is proposed Forthe latter a dual-based decomposition technique is enforcedto determine the workload of different tasks for each userWang et al [207] propose a MCS task allocation that alignstask sequences with citizensrsquo mobility By leveraging therepetition of mobility patterns the task allocation problemis studied as a pattern matching problem and the optimizationaims to pattern matching length and support degree metricsFair task allocation and energy efficiency are the main focusof [212] whose objective is to provide min-max aggregatesensing time The problem is NP-hard and it is solved with apolynomial-time approximation algorithm (for the offline case)and with a polynomial-time greedy algorithm (for the onlinecase) With Crowdlet [213] the demands for sensing data ofusers called requesters are satisfied by tasking other users inthe neighborhood for a fast response The model determinesexpected service gain that a requester can experience whenrecruiting another user The problem then turns into how tomaximize the expected sum of service quality and Crowdletproposes an optimal worker recruitment policy through thedynamic programming principle

Proper workload distribution and balancing among theparticipants is important In [206] the authors investigate thetrade-off between load balance and data utility maximizationby modeling workload allocation as a Nash bargaining game todetermine the workload of each smartphone The heterogeneousMulti-Task Assignment (HMTA) problem is presented andformalized in [214] The authors aim to maximize the quality ofinformation while minimizing the total incentive budget Theypropose a problem-solving approach of two stages by exploitingthe spatio-temporal correlations among several heterogeneous




and concurrent tasks in a shared resource pool In [215] theauthors investigate the problem of location diversity in MCSsystems where tasks arrive stochastically while citizens aremoving In the offline scenario a combinatorial algorithmefficiently allocates tasks In online scenarios the authorsinvestigate the stochastic characteristics and discontinuouscoverage to address the challenge of non-linear expectationand provide fairness among users In [216] the authors reviewthe task allocation problem by focusing on task completionin urban environments and classifying different allocationalgorithms according to related problems Often organizerswant to minimize the total costs of incentives To this enda cost-efficient incentive mechanism is presented in [217] byexploring multi-tasking versus single-task assignment scenarios

User recruitment Proper user recruitment defines the degreeof effectiveness of a MCS system hence this problem haslargely been investigated in analytical research works

In [218] the authors propose a novel dynamic user recruit-ment problem with heterogeneous sensing tasks in a large-scale MCS system They aim at minimizing the sensing costwhile satisfying a certain level of coverage in a context withdynamic tasks which change in real-time and heterogeneouswith different spatial and temporal requirements To this scopethey propose three greedy algorithms to tackle the dynamicuser recruitment problem and conduct simulations exploiting areal mobile dataset Zhang et al [219] analyze a real-world SS7data of 118 million mobile users in Xiamen China and addressa mobile user recruitment problem Given a set of target cells tobe sensed and the recruitment cost functions of the mobile usersthe MUR problem is to recruit a set of participants to cover allthe cells and minimize the total recruitment cost In [220] theauthors analyze how participants may be optimally selectedfocusing on the coverage dimension of the sensing campaigndesign problems and the complexity of opportunistic and delaytolerant networks Assuming that transferring data as soon asgenerated from the mobile devicesrsquo sensors is not the best wayfor data delivery they consider scenarios with deterministicnode mobility and propose the choice of participants as aminimum cost set cover problem with the sub-modular objectivefunction They describe practical data heuristics for the resultingNP-hard problems and in the experimentation with real mobilitydatasets provide evidence that heuristics perform better thanworst case bound suggest The authors of [221] investigateincentive mechanisms considering only the crucial dimensionof location information when tasks are assigned to users TRAC(Truthful Auction for Location-Aware Collaborative sensing) isa mechanism that is based on a reverse auction framework andconsists of a near-optimal approximate algorithm and a criticalpayment scheme CrowdRecruiter [222] and CrowdTasker [223]aim to select the minimal set of participants under probabilisticcoverage constraints while maximizing the total coverage undera fixed budget They are based on a prediction algorithm thatcomputes the coverage probability of each participant andthen exploits a near optimal function to measure the jointcoverage probability of multiple users In [224] the authorspropose incentive mechanisms that can be exploited to motivateparticipants in sensing campaigns They consider two different

approaches and propose solutions accordingly In the firstcase the responsible for the sensing campaign provides areward to the participants which they model as a Stackelberggame in which users are the followers In the other caseparticipants directly ask an incentive for their service forwhich the authors design an auction mechanism called IMCUActiveCrowd is a user selection framework for multi-task MCSarchitectures [225] The authors analyze the problem undertwo situations usersrsquo selection based on intentional individualsrsquomovement for time-sensitive tasks and unintentional movementfor delay-tolerant tasks In order to accomplish time-sensitivetasks users are required to move to the position of the taskintentionally and the aim is to minimize the total distance In thecase of delay-tolerant tasks the framework chooses participantsthat are predicted to pass close to the task and the aim is tominimize the number of required users The authors proposeto solve the two problems of minimization exploiting twoGreedy-enhanced genetic algorithms A method to optimallyselect users to generate the required space-time paths acrossthe network for collecting data from a set of fixed locations isproposed in [226] Finally one among the first works proposingincentives for crowdsourcing considers two incentive methodsa platform-centric model and a user-centric model [227] In thefirst one the platform supplies a money contribution shared byparticipants by exploiting a Stackelberg game In the secondone users have more control over the reward they will getthrough an auction-based incentive mechanism

Context awareness MCS systems are challenged by thelimited amount of resources in mobile devices in termsof storage computational and communication capabilitiesContext-awareness allows to react more smartly and proactivelyto changes in the environment around mobile devices [228] andinfluences energy and delay trade-offs of MCS systems [229]In [230] the authors focus on the trade-offs between energyconsumption due to continuous sensing and sampling rate todetect contextual events The proposed Viterbi-based Context-Aware Mobile Sensing mechanism optimizes sensing decisionsby adaptively deciding when to trigger the sensors For context-based mobile sensing in smart building [231] exploits lowpower sensors and proposes a model that enhances commu-nication data processing and analytics In [232] the authorspropose a multidimensional context-aware social networkarchitecture for MCS applications The objective is to providecontext-aware solutions of environmental personal and socialinformation for both customer and contributing users in smartcities applications In [233] the authors propose a context-aware approximation algorithm to search the optimal solutionfor a minimum vertex cover NP-complete problem that istailored for crowdsensing task assignment Understanding thephysical activity of users while performing data collection isas-well-important as context-awareness To this end recentworks have shown how to efficiently analyze sensor datato recognize physical activities and social interactions usingmachine learning techniques [234] [235]

Budget-constrained In MCS systems data collection isperformed by non-professional users with low-cost sensorsConsequently the quality of data cannot be guaranteed To




obtain effective results organizers typically reward usersaccording to the amount and quality of contributed data Thustheir objective is to minimize the budget expenditure whilemaximizing the amount and quality of gathered data

BLISS (Budget LImited robuSt crowdSensing) [112] isan online learning algorithm that minimizes the differencebetween the achieved total revenue and the optimal one It isasymptotically as the minimization is performed on averagePrediction-based user recruitment strategy in [237] dividescontributing users in two groups namely pay as you go(PAYG) and pay monthly (PAYM) Different price policiesare proposed in order to minimize the data uploading costThe authors of [236] consider only users that are located in aspace-temporal vicinity of a task to be recruited The aim is tomaximize the number of accomplished tasks under the budgetconstraint Two approaches to tackle the problem are presentedIn the first case the budget is given for a fixed and short periodwhile in the second case it is given for the entire campaignThe authors show that both variants are NP-hard and presentdifferent heuristics and a dynamic budget allocation policyin the online scenario ALSense [238] is a distributed activelearning framework that minimizes the prediction errors forclassification-based mobile crowdsensing tasks subject to dataupload and query cost constraints ABSee is an auction-basedbudget feasible mechanism that aims to maximize the quality ofsensing of users [239] Its design is based on a greedy approachand includes winner selection and payment determination rulesAnother critical challenge for campaign organizers is to set aproper posted price for a task to be accomplished with smalltotal payment and sufficient quality of data To this end [240]proposes a mechanism based on a binary search to model aseries of chance constrained posted pricing problems in robustMCS systems

D Platforms Simulators and Datasets

In the previous part of this Section we presented MCS as alayered architecture illustrated operation of DCFs to acquiredata from citizens and analytic works that address issues ofMCS systems theoretically Now we discuss practical researchworks As shown in Table VI we first present platforms fordata collection that provide resources to store process andvisualize data Then we discuss existing simulators that focuson different aspects of MCS systems Finally we analyze someexisting and publicly available collected datasets

While the main objective of platforms is to enable researchersto experiment applications in real-world simulators are bettersuited to test the technical performance of specific aspect of aMCS system for example the techniques for data reportingSimulators have the advantage of operating at large scale at theexpense of non-realistic settings while platforms are typicallylimited in the number of participants Datasets are collectedthrough real-world experiments that involve participants tosense and contribute data typically through a custom developedmobile application The importance of datasets lies in the factthat enable researchers to perform studies on the data or toemploy as inputs for simulators

Platforms Participact is a large-scale crowdsensing platformthat allows the development and deployment of experimentsconsidering both mobile device and server side [242] Itconsiders not only technical issues but also human resourcesand their involvement APISENSE is a popular cloud-basedplatform that enables researchers to deploy crowdsensingapplications by providing resources to store and process dataacquired from a crowd [243] It presents a modular service-oriented architecture on the server-side infrastructure that allowsresearchers to customize and describe requirements of experi-ments through a scripting language Also it makes available tothe users other services (eg data visualization) and a mobileapplication allowing them to download the tasks to executethem in a dedicated sandbox and to upload data on the serverautomatically SenseMyCity is a MCS platform that consists ofan infrastructure to acquire geo-indexed data through mobiledevicesrsquo sensors exploiting a multitude of voluntary userswilling to participate in the sensing campaign and the logisticsupport for large-scale experiments in urban environments It iscapable of handling data collected from different sources suchas GPS accelerometer gyroscope and environmental sensorseither embedded or external [244] CRATER is a crowdsensingplatform to estimate road conditions [245] It provides APIsto access data and visualize maps in the related applicationMedusa is a framework that provides high-level abstractionsfor analyzing the steps to accomplish a task by users [246]It exploits a distributed runtime system that coordinates theexecution and the flow control of these tasks between usersand a cluster on the cloud Prism is a Platform for RemoteSensing using Smartphones that balances generality securityand scalability [247] MOSDEN is a Mobile Sensor DataEngiNe used to capture and share sensed data among distributedapplications and several users [63] It is scalable exploitscooperative opportunistic sensing and separates the applicationprocessing from sensing storing and sharing Matador is aplatform that focuses on energy-efficient task allocation andexecution based on users context awareness [248] To this endit aims to assign tasks to users according to their surroundingpreserving the normal smartphone usage It consists of a designand prototype implementation that includes a context-awareenergy-efficient sampling algorithm aiming to deliver MCStasks by minimizing energy consumption

Simulators The success of a MCS campaign relies on alarge participation of citizens [255] Hence often it is notfeasible to deploy large-scale testbeds and it is preferableto run simulations with a precise set-up in realistic urbanenvironments Currently the existing tools aim either to wellcharacterize and model communication aspects or definethe usage of spatial environment [256] CrowdSenSim is anew tool to assess the performance of MCS systems andsmart cities services in realistic urban environments [68]For example it has been employed for analysis of city-widesmart lighting solutions [257] It is specifically designed toperform analysis in large scale environments and supports bothparticipatory and opportunistic sensing paradigms This customsimulator implements independent modules to characterize theurban environment the user mobility the communication and




TABLE VIPLATFORMS SIMULATORS AND DATASETS

WORKS DESCRIPTION REFERENCE

PLATFORMS ParticipAct Living Lab It is a large-scale crowdsensing platform that allows the development anddeployment of experiments considering both mobile device and server side

[242]

APISENSE It enables researchers to deploy crowdsensing applications by providing resourcesto store and process data acquired from a crowd

[243]

SenseMyCity It acquires geo-tagged data acquired from different mobile devicesrsquo sensors ofusers willing to participate in experiments

[244]

CRATER It provides APIs to access data and visualize maps in the related application toestimate road conditions

[245]

Medusa It provides high level abstractions for analyzing the required steps to accomplisha task by users

[246]

PRISM Platform for Remote Sensing using Smartphones that balances generality securityand scalability

[247]

MOSDEN It is used to capture and share sensed data among distributed applications andseveral users

[63]

MATADOR It aims to efficiently deliver tasks to users according to a context-aware samplingalgorithm that minimizes energy consumption of mobile devices

[248]

SIMULATORS CrowdSenSim It simulates MCS activities in large-scale urban environments implementingDCFs and realistic user mobility

[68]

NS-3 Used in a MCS environment considering mobility properties of the nodes andthe wireless interface in ad-hoc network mode

[66]

CupCarbon Discrete-event WSN simulator for IoT and smart cities which can be used forMCS purposes taking into account users as mobile nodes and base stations

[249]

Urban parking It presents a simulation environment to investigate performance of MCSapplications in an urban parking scenario

[250]

DATASETS ParticipAct It involves in MCS campaigns 173 students in the Emilia Romagna region(Italy) on a period of 15 months using Android smartphones

[64]

Cambridge It presents the mobility of 36 students in the Cambridge University Campus for12 days

[251]

MIT It provides the mobility of 94 students in the MIT Campus (Boston MA) for246 days

[252]

MDC Nokia It includes data collected from 185 citizens using a N95 Nokia mobile phonein the Lake Geneva region in Switzerland

[253]

CARMA It consists of 38 mobile users in a university campus over several weeks usinga customized crowdsourcing Android mobile application

[254]

the crowdsensing inputs which depend on the applicationand specific sensing paradigm utilized CrowdSenSim allowsscientists and engineers to investigate the performance ofthe MCS systems with a focus on data generation andparticipant recruitment The simulation platform can visualizethe obtained results with unprecedented precision overlayingthem on city maps In addition to data collection performancethe information about energy spent by participants for bothsensing and reporting helps to perform fine-grained systemoptimization Network Simulator 3 (NS-3) has been usedfor crowdsensing simulations to assess the performance ofa crowdsensing network taking into account the mobilityproperties of the nodes together with the wireless interface inad-hoc network mode [66] Furthermore the authors presenta case study about how participants could report incidents inpublic rail transport NS-3 provides highly accurate estimationsof network properties However having detailed informationon communication properties comes with the cost of losingscalability First it is not possible to simulate tens of thousandsof users contributing data Second the granularity of theduration of NS-3 simulations is typically in the order of minutesIndeed the objective is to capture specific behaviors such asthe changes in the TCP congestion window However the

duration of real sensing campaigns is typically in the order ofhours or days CupCarbon is a discrete-event wireless sensornetwork (WSN) simulator for IoT and smart cities [249] Oneof the major strengths is the possibility to model and simulateWSN on realistic urban environments through OpenStreetMapTo set up the simulations users must deploy on the map thevarious sensors and the nodes such as mobile devices andbase stations The approach is not intended for crowdsensingscenarios with thousands of users In [250] the authorspresent a simulation environment developed to investigate theperformance of crowdsensing applications in an urban parkingscenario Although the application domain is only parking-based the authors claim that the proposed solution can beapplied to other crowdsensing scenarios However the scenarioconsiders only drivers as a type of users and movementsfrom one parking spot to another one The authors considerhumans as sensors that trigger parking events However tobe widely applicable a crowdsensing simulator must considerdata generated from mobile and IoT devicesrsquo sensors carriedby human individuals

Datasets In literature it is possible to find different valuabledatasets produced by research projects which exploit MCS




platforms In these projects researchers have collected data invarious geographical areas through different sets of sensors andwith different objectives These datasets are fundamental forgiving the possibility to all the research community providinga way to test assess and compare several solutions for a widerange of application scenarios and domains The ParticipActLiving Lab is a large-scale experiment of the University ofBologna involving in crowdsensing campaigns 173 students inthe Emilia Romagna region (Italy) on a period of 15 monthsusing Android smartphones [64] [258] In this work theauthors present a comparative analysis of existing datasetsand propose ParticipAct dataset to assess the performanceof a crowdsensing platform in user assignment policies taskacceptance and task completion rate The MDC Nokia datasetincludes data collected from 185 citizens using a N95 Nokiamobile phone in the Lake Geneva region in Switzerland [253]An application running on the background collects data fromdifferent embedded sensors with a sampling period of 600 sThe Cambridge dataset presents the mobility of 36 studentsin the Cambridge University Campus for 12 days [251] Theonly information provided concerns traces of-location amongparticipants which are taken through a Bluetooth transceiverin an Intel iMote The MIT dataset provides the mobility of94 students in the MIT Campus for 246 days [252] The usersexploited an application that took into account co-locationwith other participants through Bluetooth and other databut no sensor data CARMA is a crowdsourcing Androidapplication used to perform an experimental study and collecta dataset used to derive empirical models for user mobilityand contact parameters such as frequency and duration of usercontacts [254] It has been collected through two stages the firstwith 38 mobile users contributing for 11 weeks and the secondwith 13 students gathering data over 4 weeks In additionparticipants filled a survey to investigate the correlation ofmobility and connectivity patterns with social network relations

E MCS as a Business Paradigm

Data trading has recently attracted a remarkable and increas-ing attention The analysis of information acquisition from thepoint of view of data markets is still in its infancy for bothindustry and research In this context MCS acts as a businessparadigm where citizens are at the same time data contributorscustomers and service consumers This eco-system requiresnew design strategies to enforce efficiency of MCS systems

VENUS [259] is a profit-driVEN data acqUiSition frameworkfor crowd-sensed data markets that aims to provide crowd-sensed data trading format determination profit maximizationwith polynomial computational complexity and paymentminimization in strategic environments How to regulate thetransactions between users and the MCS platform requires fur-ther investigation To give some representative examples [260]proposes a trusted contract theory-based mechanism to providesensing services with incentive schemes The objective is on theone hand to guarantee the quality of sensing while maximizingthe utility of the platform and on the other hand to satisfy userswith rewards A collaboration mechanism based on rewardingis also proposed in [261] where the organizer announces a

total reward to be shared between contributors A design ofdata sharing market and a generic pricing scheme are presentedin [262] where contributors can sell their data to customersthat are not willing to collect information on their own Inthis work the authors prove the existence and uniqueness ofthe market equilibrium and propose a quality-aware P2P-basedMCS architecture for such a data market In addition theypresent iterative algorithms to reach the equilibrium and discusshow P2P data sharing can enhance social welfare by designinga model with low trading prices and high transmission costsData market in P2P-based MCS systems is also studied in [263]that propose a hybrid pricing mechanism to reward contributors

When the aggregated information is made available as apublic good a system can also benefit from merging datacollection with routing selection algorithms through incentivemechanisms As the utility of collected information increaseswith the diversity of usersrsquo paths it is crucial to rewardparticipants accordingly For example users that move apartfrom the target path contributing data should receive a higherreward than users who do not To this end in [264] proposestwo different rewarding mechanisms to tradeoff path diversityand user participation motivating a hybrid mechanism toincrease social welfare In [265] the authors investigate theeconomics of delay-sensitive MCS campaigns by presentingan Optimal Participation and Reporting Decisions (OPRD)algorithm that aims to maximize the service providerrsquos profitIn [266] the authors study market behaviors and incentivemechanisms by proposing a non-cooperative game under apricing scheme for a theoretical analysis of the social welfareequilibrium

The cost of data transmission is one of the most influencingfactors that lower user participation To this end a contract-based incentive framework for WiFi community networksis presented in [267] where the authors study the optimalcontract and the profit gain of the operator The operatorsaugment their profit by increasing the average WiFi accessquality Although counter-intuitive this process lowers pricesand subscription fees for end-users In [268] the authors discussMCS for industrial applications by highlighting benefits andshortcomings In this area MCS can provide an efficient cost-effective and scalable data collection paradigm for industrialsensing to boost performance in connecting the componentsof the entire industrial value chain

F Final RemarksWe proposed to classify MCS literature works in a four-

layered architecture The architecture enables detailed cate-gorization of all areas of MCS from sensing to the applica-tion layer including communication and data processing Inaddition it allows to differentiate between domain-specificand general-purpose frameworks for data collection We havediscussed both theoretical works including those pertainingto the areas of operational research and optimization (eghow to maximize accomplished tasks under budget constraints)and practical ones such as platforms simulators and existingdatasets collected by research projects In addition we havediscussed MCS as a business paradigm where data is tradedas a good




ApplicationLayer

Task

User

Data LayerManagement

Processing

CommunicationLayer

Technologies

Reporting

Sensing LayerSamplingProcess

Elements

Fig 9 Taxonomies on MCS four-layered architecture It includes sensingcommunication data and application layers Sensing layer is divided betweensampling and elements which will be described in Sec VII Communicationlayer is divided between technologies and reporting which will be discussedin Sec VI Data layer is divided between management and processing andwill be presented in Sec V Application layer is divided between task anduser which will be discussed in Sec IV

In the next part of our manuscript we take the organizationof the vast literature on MCS one step further by proposingnovel taxonomies that build on and expand the discussion onthe previously introduced layered architecture In a nutshellwe i) propose new taxonomies by subdividing each layer intotwo parts as shown in Fig 9 ii) classify the existing worksaccording to the taxonomies and iii) exemplify the proposedtaxonomies and classifications with relevant cases The ultimateobjective is to classify the vast amount of still uncategorizedresearch works and establish consensus on several aspects orterms currently employed with different meanings For theclassification we resort to classifying a set of relevant researchworks that we use to exemplify the taxonomies Note that forspace reasons we will not introduce beforehand these workswith a detailed description

The outline of the following Sections is as follows Theapplication layer composed of task and user taxonomies ispresented in Sec IV The data layer composed of managementand processing taxonomies is presented in Sec V Thecommunication layer composed of technologies and reportingtaxonomies is presented in Sec VI Finally the sensing layercomposed of sampling process and elements taxonomies willbe discussed in Sec VII

IV TAXONOMIES AND CLASSIFICATION ONAPPLICATION LAYER

This Section analyzes the taxonomies of the application layerThis layer is mainly composed by task and user categorieswhich are discussed with two corresponding taxonomies(Subsection IV-A) Then we classify papers accordingly(Subsection IV-B) Fig 10 shows the taxonomies on theapplication layer

A Taxonomies

1) Task The upper part of Fig 10 shows the classificationof task-related taxonomies which comprise pro-active andreactive scheduling methodologies for task assignment andexecution

Scheduling Task scheduling describes the process of allocatingthe tasks to the users We identify two strategies for theassignment on the basis of the behavior of a user Withproactive scheduling users that actively sense data without apre-assigned task eg to capture a picture Reactive schedulingindicates that users receive tasks and contribute data toaccomplish the received jobs

Proactive Under pro-active task scheduling users can freelydecide when where and when to contribute This approachis helpful for example in the public safety context to receiveinformation from users that have assisted to a crash accidentSeveral social networks-based applications assign tasks onlyafter users have already performed sensing [59] [149]

Reactive Under reactive task scheduling a user receives arequest and upon acceptance accomplishes a task Tasksshould be assigned in a reactive way when the objective to beachieved is already clear before starting the sensing campaignTo illustrate with an example monitoring a phenomenon likeair pollution [56] falls in this category

Assignment It indicates how tasks are assigned to usersAccording to the entity that dispatches tasks we classifyassignment processes into centralized where there is a centralauthority and decentralized where users themselves dispatchtasks [15]

Central authority In this approach a central authority dis-tributes tasks among users Typical centralized task distributionsinvolve environmental monitoring applications such as detec-tion of ionosphere pollution [128] or nuclear radiation [120]

Decentralized When the task distribution is decentralized eachparticipant becomes an authority and can either perform thetask or to forward it to other users This approach is very usefulwhen users are very interested in specific events or activitiesA typical example is Mobile Social Network or IntelligentTransportation Systems to share news on public transportdelays [142] or comparing prices of goods in real-time [133]

Execution This category defines the methodologies for taskexecution

Single task Under this category fall those campaigns whereonly one type of task is assigned to the users for instancetaking a picture after a car accident or measuring the level ofdecibel for noise monitoring [123]

Multi-tasking On the contrary a multi-tasking campaignexecution foresees that the central authority assigns multipletypes of tasks to users To exemplify simultaneous monitoringof noise and air quality

2) User The lower part of Fig 10 shows the user-relatedtaxonomies that focus on recruitment strategy selection criteriaand type of users




Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric

Fig 10 Taxonomies on application layer which is composed of task and user categories The task-related taxonomies are composed of scheduling assignmentand execution categories while user-related taxonomies are divided into recruitment type and selection categories

Recruitment Citizen recruitment and data contribution incen-tives are fundamental to the success of MCS campaigns Inliterature the term user recruitment is used with two differentmeanings The first and the most popular is related to citizenswho join a sensing campaign and become potential contributorsThe second meaning is less common but is employed in morerecent works and indicates the process of selecting users toundertake a given task from the pool of all contributors Tocapture such a difference we introduce and divide the termsuser recruitment and user selection as illustrated in Fig11 Inour taxonomy user recruitment constitutes only the processof recruiting participants who can join on a voluntary basisor through incentives Then sensing tasks are allocated tothe recruited users according to policies specifically designedby the sensing campaign organizer and contributing users areselected between them We will discuss user selection in thefollowing paragraph

The strategies to obtain large user participation are strictlyrelated to the type of application For example for a health-care application that collects information on food allergies(or gluten-free [116]) people affected by the problem usuallyvolunteer to participate However if the application target ismore generic and does not match well with usersrsquo interests(eg noise monitoring [123]) the best solution is to encourageuser participation through incentive mechanisms It shouldbe noted that these strategies are not mutually exclusive andusers can first volunteer and then be rewarded for qualityexecution of sensing tasks Such a solution is well suited forcases when users should perform additional tasks (eg sendingmore accurate data) or in particular situations (eg few userscontribute from a remote area)

Voluntary participation Citizens can volunteer to join a sensingcampaign for personal interests (eg when mapping restaurantsand voting food for celiacs or in the healthcare) or willingnessto help the community (eg air quality and pollution [56]) Involunteer-built MCS systems users are willing to participateand contribute spontaneously without expecting to receive

Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc

User SelectionFig 11 User recruitment and user selection process The figure shows thatusers are recruited between citizens then tasks are tentatively allocated to userswhich can accept or not Upon acceptance users are selected to contributedata to the central collector

monetary compensation Apisense is an example of a platformhelping organizers of crowdsensing systems to collect datafrom volunteers [269]Recruitment through incentives To promote participation andcontrol the rate of recruitment a set of user incentives can beintroduced in MCS [227] [270]ndash[272] Many strategies havebeen proposed to stimulate participation [11] [15] [273] Itis possible to classify the existing works on incentives intothree categories [12] entertainment service and money Theentertainment category stimulates people by turning somesensing tasks into games The service category consists inrewarding personal contribution with services offered by thesystem Finally monetary incentives methods provide money asa reward for usersrsquo contribution In general incentives shouldalso depend on the quality of contributed data and decided onthe relationship between demand and supply eg applyingmicroeconomics concepts [274] MCS systems may considerdistributing incentives in an online or offline scenario [275]

Selection User selection consists in choosing contributors to




a sensing campaign that better match its requirements Such aset of users is a subset of the recruited users Several factorscan be considered to select users such as temporal or spatialcoverage the density of users in certain regions of interest orvariable user availability We mainly classify the process ofuser selection through the classes user centric and platformcentric

User centric When the selection is user centric contributionsdepend only on participants willingness to sense and report datato the central collector which is not responsible for choosingthem

Platform centric User selection is platform centric when thecentral authority directly decides data contributors The decisionis usually based on parameters decided by organizers such astemporal and spatial coverage the total amount of collecteddata mobility of users or density of users and data in aparticular region of interests In other words platform centricdecisions are taken according to the utility of data to accomplishthe targets of the campaign

Type This category classifies users into contributors andconsumers Contributors are individuals that sense and reportdata to a collector while the consumers utilize sensed datawithout having contributed A user can also join a campaignassuming both roles For instance users can send pictures withthe price of fuel when they pass by a gas station (contributors)while at the same time the service shows fuel prices at thenearby gas stations (consumer) [131]

Contributor A contributor reports data to the MCS platformwith no interest in the results of the sensing campaignContributors are typically driven by incentives or by the desireto help the scientific or civil communities (eg a person cancollect data from the microphone of a mobile device to mapnoise pollution in a city [115] with no interests of knowingresults)

Consumer Citizens who join a service to obtain informationabout certain application scenario and have a direct interestin the results of the sensing campaign are consumers Celiacpeople for instance are interested in knowing which restaurantsto visit [116]

B Classification

This part classifies and surveys literature works accordingto the task- and user-related taxonomies previously proposedin this Section

Task The following paragraphs survey and classify researchworks according the task taxonomies presented in IV-A1Table VII illustrates the classification per paper

Scheduling A user contributes data in a pro-active way accord-ing to hisher willingness to sense specific events (eg taking apicture of a car accident) Typical pro-active applications are inhealth care [17] [19] [136] and mobile social networks [59][60] [151] [152] [155] domains Other examples are sharingprices of goods [132] [133] and recommending informationto others such as travelling advice [169] Reactive schedulingconsists in sending data according to a pre-assigned task

such as traffic [20] conditions detection air quality [21] andnoise monitoring [115] [121]ndash[123] Other examples consistin conferences [167] emotional situations [149] detectingsounds [166] characterizing WiFi coverage [164] spaceweather [129] While in MSNs applications are typically areactive scheduling Whozthat [153] and MobiClique [150]represent two excpetions

Assignment Task assignment can be centralized or decentralizedaccording to the entity that performs the dispatch Tasks aretypically assigned by a central authority when data aggrega-tion is needed to infer information such as for monitoringtraffic [20] [145] conditions air pollution [21] noise [115][122] [123] [166] and weather [129] Tasks assignment isdecentralized when users are free to distribute tasks amongthemselves in a peer-to-peer fashion for instance sendingpictures of a car accident as alert messages [143] Typicaldecentralized applications are mobile social networks [59][60] [60] [150]ndash[153] or sharing info about services [132][133] [169]

Execution Users can execute one or multiple tasks simultane-ously in a campaign Single task execution includes all MCScampaigns that require users to accomplish only one tasksuch as mapping air quality [21] or noise [115] [122] [123]and detecting prices [132] [133] The category multi-taskingspecifies works in which users can contribute multiple tasks atthe same time or in different situations For instance mobilesocial networks imply that users collect videos images audiosrelated to different space and time situations [59] [60] [60][150] [155] Healthcare applications usually also require amultitasking execution including different sensors to acquiredifferent information [17] [19] [20] such as images videosor physical activities recognized through activity pattern duringa diet [136]

User The following paragraphs survey and classify researchworks according to the user-related taxonomies on recruitmentand selection strategies and type of users presented in IV-A2Table VIII shows the classification per paper

Recruitment It classifies as voluntary when users do notreceive any compensation for participating in sensing andwhen they are granted with compensation through incentivemechanisms which can include monetary rewarding services orother benefits Typical voluntary contributions are in health careapplications [17] [19] [136] such as celiacs who are interestedin checking and sharing gluten-free food and restaurants [116]Users may contribute voluntary also when monitoring phenom-ena in which they are not only contributors but also interestedconsumers such as checking traffic and road conditions [20][142] [143] [145] environment [122] [123] or comparingprices of goods [131] [132] Mobile social networks are alsobased on voluntary interactions between users [59] [60] [121][150] [151] [153] [155] Grant users through incentives is thesimplest way to have a large amount of contributed data Mostcommon incentive is monetary rewarding which is neededwhen users are not directly interested in contributing forinstance monitoring noise [115] air pollution [21] and spaceweather [129]




TABLE VIICLASSIFICATION BASED ON TASK TAXONOMIES OF APPLICATION LAYER

SCHEDULING ASSIGNMENT EXECUTION

PROJECT REFERENCE Pro-active Reactive Central Aut Decentralized Single task Multi-tasking

HealthAware [17] x x xDietSense [19] x x xNericell [20] x x xNoiseMap [115] x x xGasMobile [21] x x xNoiseTube [123] x x xCenceMe [59] x x xMicroBlog [60] x x xPEIR [121] x x xHow long to wait [142] x x xPetrolWatch [131] x x xAndWellness [136] x x xDarwin [155] x x xCrowdSensePlace [147] x x xILR [168] x x xSoundSense [166] x x xUrban WiFi [164] x x xLiveCompare [133] x x xMobiClique [150] x x xMobiShop [132] x x xSPA [134] x x xEmotionSense [149] x x xConferenceSense [167] x x xTravel Packages [169] x x xMahali [129] x x xEar-Phone [122] x x xWreckWatch [143] x x xVTrack [145] x x xSocial Serendipity [151] x x xSociableSense [152] x x xWhozThat [153] x x xMoVi [154] x x x

Selection It overviews works between user or platform centricselection User selection is based on contributors willingnessto sense and report data Voluntary-based applications withinterested contributors typically employ this approach suchas mobile social networks [59] [60] [155] comparing livepricing [131]ndash[133] wellbeing [17] [136] and monitoringtraffic [20] [142] Platform centric selection is used whenthe central authority requires a specific sensing coverage ortype of users To illustrate with few examples in [276] theorganizers can choose well-suited participants according tospecific features such as their habits and data collection is basedon their geographical and temporal availability In [61] theauthors propose a geo-social model that selects the contributorsaccording to a matching algorithm Other frameworks extendthe set of criteria for recruitment [277] including parameterssuch as the distance between a sensing task and users theirwillingness to perform the task and remaining battery lifeNericell [20] collects data of road and traffic condition fromspecific road segments in Bangalore while Gas Mobile [21]took measurements of air quality from a set of bicycles movingaround several bicycle rides all around the considered city

Type It classifies users between consumers who use a serviceprovided by MCS organizer (eg sharing information aboutfood [116]) and contributors who sense and report data withoutusing the service In all the works we considered users behavemainly as contributors because they collect data and deliverit to the central authority In some of the works users mayalso act as customers In health care applications users aretypically contributors and consumers because they not onlyshare personal data but also receive a response or informationfrom other users [17] [19] [136] Mobile social networks alsoexploit the fact that users receive collected data from otherparticipants [19] [60] [121] [151] [153] [155] E-commerceis another application that typically leverages on users that areboth contributors and consumers [131]ndash[133]

V TAXONOMIES AND CLASSIFICATION ON DATA LAYER

This section analyzes the taxonomies on the data layerand surveys research works accordingly Data layer comprisesmanagement and processing Fig 12 shows the data-relatedtaxonomies




TABLE VIIICLASSIFICATION BASED ON USER TAXONOMIES OF APPLICATION LAYER

RECRUITMENT SELECTION TYPE

PROJECT REFERENCE Voluntary Incentives Platform centric User centric Consumer Contributor

HealthAware [17] x x x xDietSense [19] x x x xNericell [20] x x xNoiseMap [115] x x xGasMobile [21] x x xNoiseTube [123] x x xCenceMe [59] x x x xMicroBlog [60] x x x xPEIR [121] x x x xHow long to wait [142] x x x xPetrolWatch [131] x x x xAndWellness [136] x x x xDarwin [155] x x x xCrowdSensePlace [147] x x xILR [168] x x xSoundSense [166] x x xUrban WiFi [164] x x xLiveCompare [133] x x x xMobiClique [150] x x x xMobiShop [132] x x x xSPA [134] x x x xEmotionSense [149] x x xConferenceSense [167] x x xTravel Packages [169] x x x xMahali [129] x x xEar-Phone [122] x x xWreckWatch [143] x x xVTrack [145] x x x xSocial Serendipity [151] x x x xSociableSense [152] x x x xWhozThat [153] x x x xMoVi [154] x x x x

A Taxonomies

1) Management Data management is related to storageformat and dimension of acquired data The upper part ofFig 12 shows the subcategories for each taxonomy

Storage The category defines the methodologies to keep andmaintain the collected data Two subcategories are defined iecentralized and distributed according to the location wheredata is made available

Centralized A MCS system operates a centralized storagemanagement when data is stored and maintained in a singlelocation which is usually a database made available in thecloud This approach is typically employed when significantprocessing or data aggregation is required For instance urbanmonitoring [140] price comparison [133] and emergency situ-ation management [118] are domains that require a centralizedstorage

Distributed Storing data in a distributed manner is typicallyemployed for delay-tolerant applications ie when users areallowed to deliver data with a delayed reporting For instancewhen labeled data can be aggregated at the end of sensing

campaign to map a phenomenon such as air quality [56] andnoise monitoring [115] as well as urban planning [162] Therecent fog computing and mobilemulti-access edge computingthat make resources available close to the end users are also anexample of distributed storage [278] We will further discussthese paradigms in Sec VIII-B

Format The class format divides data between structuredand unstructured Structured data is clearly defined whileunstructured data includes information that is not easy tofind and requires significant processing because it comes withdifferent formats

Structured Structured data has been created to be stored asa structure and analyzed It is organized to let easy accessand analysis and typically is self-explanatory Representativeexamples are identifiers and specific information such as ageand other characteristics of individuals

Unstructured Unstructured data does not have a specific identi-fier to be recognized by search functions easily Representativeexamples are media files such as video audio and images andall types of files that require complex analysis eg information




Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data

Filtering amp denoising

ML amp data mining

Real-time

Statistical

Prediction

Fig 12 Taxonomies on data layer which includes management and processing categories The management-related taxonomies are composed of storageformat and dimension classes while processing-related taxonomies are divided into pre-processing analytics and post-processing classes

collected from social networks

Dimension Data dimension is related to the set of attributes ofthe collected data sets For single dimension data the attributescomprise a single type of data and multi-dimensional whenthe data set includes more types of data

Single dimension Typically applications exploiting a specifictype of sensor produce single dimension data Representativeexamples are environmental monitoring exploiting a dedicatedsensor eg air quality or temperature mapping

Multi-dimensional Typically applications that require the useof different sensors or multimedia applications produce multi-dimensional data For instance mobile social networks areexamples of multi-dimensional data sets because they allowusers to upload video images audio etc

2) Processing Processing data is a fundamental step inMCS campaigns The lower part of Fig 12 shows the classesof processing-related taxonomies which are divided betweenpre-processing analytics and post-processing

Pre-processing Pre-processing operations are performed oncollected data before analytics We divide pre-processing intoraw data and filtering and denoising categories Data is rawwhen no operations are executed before data analytics like infiltering and denoising

Raw data Raw data has not been treated by any manipulationoperations The advantage of storing raw data is that it is alwayspossible to infer information applying different processingtechniques at a later stage

Filtering and denoising Filtering and denoising refer to themain strategies employed to refine collected data by removingirrelevant and redundant data In addition they help to aggregateand make sense of data while reducing at the same time thevolume to be stored

Analytics Data analytics aims to extract and expose meaning-ful information through a wide set of techniques Corresponding

taxonomies include machine learning (ML) amp data mining andreal-time analytics

ML and data mining ML and data mining category analyzedata not in real-time These techniques aim to infer informationidentify patterns or predict future trends Typical applicationsthat exploit these techniques in MCS systems are environmentalmonitoring and mapping urban planning and indoor navigationsystems

Real-time Real-time analytics consists in examining collecteddata as soon as it is produced by the contributors Ana-lyzing data in real-time requires high responsiveness andcomputational resources from the system Typical examplesare campaigns for traffic monitoring emergency situationmanagement or unmanned vehicles

Post-processing After data analytics it is possible to adoptpost-processing techniques for statistical reasons or predictiveanalysis

Statistical Statistical post-processing aims at inferring propor-tions given quantitative examples of the input data For examplewhile mapping air quality statistical methods can be applied tostudy sensed information eg analyzing the correlation of rushhours and congested roads with air pollution by computingaverage values

Prediction Predictive post-processing techniques aim at de-termining future outcomes from a set of possibilities whengiven new input in the system In the application domain of airquality monitoring post-processing is a prediction when givena dataset of sensed data the organizer aims to forecast whichwill be the future air quality conditions (eg Wednesday atlunchtime)

B Classification

In the following paragraphs we survey literature worksaccording to the taxonomy of data layer proposed above




TABLE IXCLASSIFICATION BASED ON MANAGEMENT TAXONOMY OF DATA LAYER

STORAGE FORMAT DIMENSION

PROJECT REFERENCE Centralized Distributed Structured Unstructured Single dimension Multi-dimensional


Management The research works are classified and surveyedin Table IX according to data management taxonomies dis-cussed in V-A1

Storage The centralized strategy is typically used when thecollector needs to have insights by inferring information fromraw data or when data needs to be aggregated To giverepresentative examples the centralized storage is exploitedin monitoring noise [115] [123] air quality [21] [22] [56]traffic conditions [20] [142] prices of goods [131] [133]MCS organizers exploit a distributed approach when they donot need to aggregate all gathered data or it is not possiblefor different constraints such as privacy reasons In order toaddress the integrity of the data and the contributorsrsquo privacythe CenceMe application stores locally data to be processedby the phone classifiers and none of the raw collected datais sent to the backend [59] For same reasons applicationsrelated to healthcare [17] sharing travels [169] emotions [149]sounds [166] [167] or locations [164] [168] adopt a distributedlocal storage

Format Structured data is stored and analyzed as a singleand self-explanatory structure which is easy to be aggregatedand compare between different contributions For instancesamples that are aggregated together to map a phenomenon are

structured data Representative examples consist of environ-mental monitoring [21] [115] [122] [123] checking pricesin real-time [131] [133] characterizing WiFi intensity [164]Unstructured data cannot be compared and do not have aspecific value being characterized by multimedia such asmobile social applications [150] [151] [155] comparing travelpackages [169] improving location reliability [168] estimatingbus arrival time [142] and monitoring health conditions [17][136]

Dimension Single dimension approach is typical of MCSsystems that require data gathered from a specific sensorTypical examples are monitoring phenomena that couple adetected value with its sensing location such as noise [115][122] [123] and air quality [21] monitoring mapping WiFiintensity [164] and checking prices of goods [131] [150]Multi-dimensional MCS campaigns aim to gather informationfrom different types of sensors Representative examples aremobile social networks [60] [152] [153] [155] and applica-tions that consist in sharing multimedia such as travelling [169]conferences [167] emotions [149]

Processing Here we survey and classify works according theprocessing taxonomy presented in V-A2 Surveyed papers andtheir carachteristics are shown in Table X




TABLE XCLASSIFICATION BASED ON PROCESSING TAXONOMY OF DATA LAYER

PRE-PROCESSING ANALYTICS POST-PROCESSING

PROJECT REFERENCE Raw data Filtering amp denoising ML amp data mining Real-time Statistical Prediction


Pre-processing It divides raw data and filtering amp denoisingcategories Some applications collect raw data without anyfurther data pre-processing In most cases works that exploitraw data simply consider different types of collected data suchas associating a position taken from the GPS with a sensedvalue such as dB for noise monitoring [115] [122] [123] apower for WiFi strength [164] a price for e-commerce [132][133] Filtering amp denoising indicates applications that performoperations on raw data For instance in health care applicationsis possible to recognize the activity or condition of a userfrom patterns collected from different sensors [17] suchas accelerometer or gyroscope Other phenomena in urbanenvironments require more complex pre-processing operationssuch as detecting traffic [20] [143]

Analytics Most of MCS systems perform ML and data miningtechniques after data is collected and aggregated Typicalapplications exploiting this approach consist in mappingphenomena and resources in cities such as WiFi intensity [164]air quality [21] [22] [56] noise [115] [123] Other appli-cation domains employing this approach are mobile socialnetworks [59] [60] [154] and sharing information such astravel packages [169] food [116] sounds [166] and improvinglocation reliability [168] Darwin phones [155] performs ML

techniques for privacy constraints aiming to not send to thecentral collector personal data that are deleted after beinganalyzed in the mobile device which sends only inferredinformation Real-time analytics aim to infer useful informationas soon as possible and require more computational resourcesConsequently MCS systems employ them only when needed byapplication domains such as monitoring traffic conditions [20]waiting time of public transports [142] comparing prices [131][133] In WhozThat [153] is an exception of mobile socialnetworks where real-time analytics is performed for context-aware sensing and detecting users in the surroundings In healthcare AndWellness [136] and SPA [134] have data mininganalytics because it aims to provide advice from remote in along-term perspective while HealtAware [17] presents real-timeanalytics to check conditions mainly of elderly people

Post-processing Almost all of the considered works presentstatistical post-processing where inferred information is ana-lyzed with statistical methods and approaches Environmentalmonitoring [21] [56] [115] [123] healthcare [17] [136]mobile social networks [59] [154] [155] are only a fewexamples that adopt statistical post-processing Considering theworks under analysis the predictive approach is exploited onlyto predict the waiting time for bus arrival [142] and estimate




traffic delays by VTrack [145] Nonetheless recent MCSsystems have adopted predictive analytics in many differentfields such as unmanned vehicles and urban planning alreadydiscussed in Sec III-B In particular urbanization issues requirenowadays to develop novel techniques based on predictiveanalytics to improve historical experience-based decisions suchas where to open a new local business or how to managemobility more smartly We will further discuss these domainsin Sec VIII-B

VI TAXONOMIES AND CLASSIFICATION ONCOMMUNICATION LAYER

This Section presents the taxonomies on the communicationlayer which is composed by technologies and reporting classesThe technologies taxonomy analyzes the types of networkinterfaces that can be used to report data The reportingtaxonomy analyzes different ways of reporting data whichare not related to technologies but depends on the applicationdomains and constraints of sensing campaigns Fig 13 showssubcategories of taxonomies on communication layer

A Taxonomies

1) Technologies Infrastructured and infrastructure-less con-stitute the subcategories of the Technologies taxonomy asshown in the upper part of Fig 13 Infrastructured technologiesare cellular or wireless local area network (WLAN) wherethe network relies on base stations or access points to estab-lish communication links In infrastructure-less technologiesproximity-based communications are enforced with peer-to-peertechnologies such as LTE-Direct WiFi-Direct and Bluetooth

Infrastructured Infrastructured technologies indicate the needfor an infrastructure for data delivery In this class we considercellular data communications and WLAN interfaces Accordingto the design of each campaign organizers can require a specifictechnology to report data or leave the choice to end usersUsually cellular connectivity is used when a sensing campaignrequires data delivery with bounds on latency that WiFi doesnot guarantee because of its unavailability in many places andcontention-based techniques for channel access (eg building areal-time map for monitoring availability of parking slot [34])On the other side WLAN interfaces can be exploited formapping phenomena that can tolerate delays and save costs toend users

Cellular Reporting data through cellular connectivity is typ-ically required from sensing campaign that perform real-time monitoring and require data reception as soon as it issensed Representative areas where cellular networking shouldbe employed are traffic monitoring unmanned vehicles andemergency situation management

From a technological perspective the required data rates andlatencies that current LTE and 4G systems provide are sufficientfor the purposes of MCS systems The upcoming 5G revolutionwill however be relevant for MCS because of the followingmain reasons First 5G architecture is projected to be highlybased on the concepts of network function virtualization andSDN As a consequence the core functionalities that now are

performed by on custom-built hardware will run in distributedfashion leading to better mobility support Second the use ofadditional frequencies for example in the millimeter-wave partof the spectrum opens up the doors for higher data rates at theexpense of high path-loss and susceptibility to blockages Thiscalls for much denser network deployments that make MCSsystems benefit from better coverage Additionally servicesrequiring higher data rates will be served by millimeter-wavebase stations making room for additional resources in thelower part of the spectrum where MCS services will be served

WLAN Typically users tend to exploit WLAN interfaces tosend data to the central collector for saving costs that cellularnetwork impose with the subscription fees The main drawbackof this approach is the unavailability of WiFi connectivity Con-sequently this approach is used mainly when sensing organizersdo not specify any preferred reporting technologies or when theapplication domain permits to send data also a certain amountof time after the sensing process Representative domains thatexploit this approach are environmental monitoring and urbanplanning

Infrastructure-less It consists of device-to-device (D2D)communications that do not require any infrastructure as anetwork access point but rather allow devices in the vicinityto communicate directly

WiFi-direct WiFi Direct technology is one of the most popularD2D communication standard which is also suitable in thecontext of MCS paradigm A WiFi Direct Group is composedof a group owner (GO) and zero or more group members(GM) and a device can dynamically assume the role of GMor GO In [279] the authors propose a system in which usersreport data to the central collector collaboratively For eachgroup a GO is elected and it is then responsible for reportingaggregated information to the central collector through LTEinterfaces

LTE-direct Among the emerging D2D communication tech-nologies LTE-Direct is a very suitable technology for theMCS systems Compared to other D2D standards LTE-Directpresents different characteristics that perfectly fit the potentialof MCS paradigm Specifically it exhibits very low energyconsumption to perform continuous scanning and fast discoveryof mobile devices in proximity and wide transmission rangearound 500 meters that allows to create groups with manyparticipants [280]

Bluetooth Bluetooth represents another energy-efficient strategyto report data through a collaboration between users [281] Inparticular Bluetooth Low Energy (BLE) is the employed lowpower version of standard Bluetooth specifically built for IoTenvironments proposed in the Bluetooth 40 specification [282]Its energy efficiency perfectly fits the usage of mobile deviceswhich require to work for long period with limited use ofresources and low battery drain

2) Reporting The lower part of Fig 13 shows taxonomiesrelated to reporting which are divided into upload modemethodology and timing classes Unlike the previous categoryreporting defines the ways in which the mobile devices report




CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous

Fig 13 Taxonomies on communication layer which comprises technologies and reporting categories The technologies-related taxonomies are composed ofinfrastructured and infrastructure-less classes while reporting-related taxonomies are divided into upload mode methodology and timing classes

sensed data to the central collector Upload mode describesif data are delivered in real-time or in a delayed mannerThe methodology category investigates how a mobile deviceexecutes the sensing process by itself and concerning otherdevices Finally timing analyzes if reporting between differentcontributors is synchronous or not

Upload mode Upload can be relay or store and forwardaccording to delay tolerance required by the applicationRelay In this method data is delivered as soon as collectedWhen mobile devices cannot meet real-time delivery the bestsolution is to skip a few samples and in turns avoiding to wasteenergy for sensing or reporting as the result may be inconsistentat the data collector [236] Typical examples of time-sensitivetasks are emergency-related or monitoring of traffic conditionsand sharing data with users through applications such asWaze [283]Store and forward This approach is typically used in delay-tolerant applications when campaigns do not need to receivedata in real-time Data is typically labeled to include areference to the sensed phenomenon [220] For instance trafficmonitoring [142] or gluten sensors to share restaurants andfood for people suffering celiac disease [116] are examples ofsensing activities that do not present temporal constraints

Methodology This category considers how a device executesa sensing process in respect to others Devices can act as peersand accomplish tasks in a collaborative way or individuallyand independently one from each otherIndividual Sensing execution is individual when each useraccomplishes the requested task individually and withoutinteraction with other participantsCollaborative The collaborative approach indicates that userscommunicate with each other exchange data and help them-selves in accomplishing a task or delivering information to thecentral collector Users are typically grouped and exchangedata exploiting short-range communication technologies suchas WiFi-direct or Bluetooth [281] Collaborative approaches

can also be seen as a hierarchical data collection process inwhich some users have more responsibilities than others Thispaves the path for specific rewarding strategies to incentiveusers in being the owner of a group to compensate him for itshigher energy costs

Timing Execution timing indicates if the devices should sensein the same period or not This constraint is fundamental forapplications that aim at comparing phenomena in a certaintime window Task reporting can be executed in synchronousor asynchronous fashion

Synchronous This category includes the cases in which usersstart and accomplish at the same time the sensing task Forsynchronization purposes participants can communicate witheach other or receive timing information from a centralauthority For instance LiveCompare compares the live price ofgoods and users should start sensing synchronously Otherwisethe comparison does not provide meaningful results [133]Another example is traffic monitoring [20]

Asynchronous The execution time is asynchronous whenusers perform sensing activity not in time synchronizationwith other users This approach is particularly useful forenvironmental monitoring and mapping phenomena receivinglabeled data also in a different interval of time Typicalexamples are mapping noise [115] and air pollution [21] inurban environments

B Classification

The following paragraphs classify MCS works accordingto the communication layer taxonomy previously described inthis Section

Technologies Literature works are classified and surveyed inTable XI according to data management taxonomies discussedin VI-A1

Infrastructured An application may pretend a specific com-munication technology for some specific design requirement




TABLE XICLASSIFICATION BASED ON TECHNOLOGIES TAXONOMY OF COMMUNICATION LAYER

INFRASTRUCTURED INFRASTRUCTURE-LESS

PROJECT REFERENCE Cellular WLAN LTE-Direct WiFi-Direct Bluetooth

HealthAware [17] x xDietSense [19] x xNericell [20] x xNoiseMap [115] x xGasMobile [21] x xNoiseTube [123] x xCenceMe [59] x xMicroBlog [60] x xPEIR [121] x x xHow long to wait [142] xPetrolWatch [131] xAndWellness [136] x xDarwin [155] x x xCrowdSensePlace [147] xILR [168] x x xSoundSense [166] x xUrban WiFi [164] xLiveCompare [133] x xMobiClique [150] x x xMobiShop [132] x xSPA [134] x xEmotionSense [149] x x xConferenceSense [167] x xTravel Packages [169] x xMahali [129] x xEar-Phone [122] x xWreckWatch [143] xVTrack [145] x xSocial Serendipity [151] x x xSociableSense [152] x x xWhozThat [153] x x xMoVi [154] x x

a Note that the set of selected works was developed much before the definition of thestandards LTE-Direct and WiFi-Direct thus these columns have no corresponding marks

or leave to the users the choice of which type of transmissionexploit Most of the considered applications do not specify arequired communication technology to deliver data Indeed theTable presents corresponding marks to both WLAN and cellulardata connectivity Applications that permit users to send data aspreferred aim to collect as much data as possible without havingspecific design constraints such as mobile social networks [59][153]ndash[155] mapping noise [115] [122] [123] or air pollution[21] [22] or monitoring health conditions [19] [136] Otherexamples consist in applications that imply a participatoryapproach that involves users directly and permit them tochoose how to deliver data such as sharing information onenvironment [121] sounds [166] emotions [149] travels [169]and space weather monitoring [129] Some applications requirea specific communication technology to deliver data for manydifferent reasons On the one side MCS systems that requiredata cellular connectivity typically aim to guarantee a certainspatial coverage or real-time reporting that WiFi availabilitycannot guarantee Some representative examples are monitoringtraffic conditions [20] [143] predicting bus arrival time [142]and comparing fuel prices [131] On the other side applicationsthat require transmission through WLAN interfaces usuallydo not have constraints on spatial coverage and real-time datadelivery but aim to save costs to users (eg energy and dataplan consumptions) For instance CrowdsensePlace [147]

SPA [134] and HealthAware [17] transmit only when a WiFiconnection is available and mobile devices are line-poweredto minimize energy consumption

Infrastructure-less Recently MCS systems are increasingthe usage of D2D communication technologies to exchangedata between users in proximity While most important MCSworks under analysis for our taxonomies and correspondingclassification have been developed before the definition of WiFi-Direct and LTE-Direct and none of them utilize these standardssome of them employ Bluetooth as shown in Table XI Forinstance Bluetooth is used between users in proximity forsocial networks purposes [151] [153] [155] environmentalmonitoring [121] and exchanging information [149] [167]

Reporting Literature works are classified and surveyed inTable XII according to data management taxonomies discussedin VI-A2

Upload mode Upload can can be divided between relay orstore amp forward according to the application requirementsRelay is typically employed for real-time monitoring suchas traffic conditions [20] [142] [143] [145] or price com-parison [132] [133] Store amp forward approach is used inapplications without strict time constraints that are typicallyrelated to labeled data For instance mapping phenomena in acity like noise [122] [123] air quality [21] or characterizing




TABLE XIICLASSIFICATION BASED ON REPORTING TAXONOMY OF COMMUNICATION LAYER

UPLOAD MODE METHODOLOGY TIMING

PROJECT REFERENCE Relay Store amp forward Individual Collaborative Synchronous Asynchronous


WiFi coverage [164] can be reported also at the end of theto save energy Other applications are not so tolerant butstill do not require strict constraints such as mobile socialnetworks [59] [60] [150]ndash[153] [155] health care [17] [19]price comparison [131] reporting info about environment [121]sounds [166] conferences [167] or sharing and recommendingtravels [169] Despite being in domains that usually aredelay tolerant NoiseMap [115] and AndWellness [136] areapplications that provide real-time services and require therelay approachMethodology It is divided between individual and collaborativeMCS systems Most of MCS systems are based on individualsensing and reporting without collaboration between the mobiledevices Typical examples are healthcare applications [17][19] noise [115] [122] [123] [136] Note that systems thatcreate maps merging data from different users are consideredindividual because users do not interact between each other tocontribute Some examples are air quality [21] [22] [56] andnoise monitoring [115] traffic estimation [20] [143] [145]Mobile social networks are typical collaborative systems beingcharacterized by sharing and querying actions that requireinteractions between users [60] [62] [151] [153] [155] To

illustrate monitoring prices in e-commerce [132] [133] and thebus arrival time [142] are other examples of user collaboratingto reach the scope of the application

Timing This classification specifies if the process of sensingneeds users contributing synchronously or not The synchronousapproach is typically requested by applications that aim atcomparing services in real-time such as prices of goods [132][133] Monitoring and mapping noise pollution is an exampleof application that can be done with both synchronous orasynchronous approach For instance [115] requires a syn-chronous design because all users perform the sensing at thesame time Typically this approach aims to infer data andhave the results while the sensing process is ongoing suchas monitoring traffic condition is useful only if performed byusers at the same time [20] Asynchronous MCS system donot require simultaneous usersrsquo contribution Mapping WiFisignal strength intensity [164] or noise pollution [123] [122] aretypically performed with this approach Healthcare applicationsdo not require contemporary contributions [17] Some mobilesocial networks do not require users to share their interests atthe same time [152] [153]




VII TAXONOMIES AND CLASSIFICATION ON SENSINGLAYER

This Section discusses the taxonomies of the sensing layerA general representation of the sensing layer is presented inFig 8 The taxonomy is composed by elements and samplingprocess categories that are respectively discussed in Subsec-tions VII-A1 and VII-A2 Then Subsection VII-B classifiesthe research works according to the proposed taxonomies ThisSection does not delve into the technicalities of sensors as thereis a vast literature on regard We refer the interested reader tosurveys on smartphone sensors [48] [49] and mobile phonesensing [8] [9]

A Taxonomies

1) Elements As shown in the upper part of Fig 14 sensingelements can be mainly classified into three characteristicsaccording to their deployment activity and acquisition In thedeployment category we distinguish between the dedicated andnon-dedicated deployment of sensors in the mobile devicesThe activity differentiates between sensors that are always-onbecause they are assigned basic operations of the device andthose that require user intervention to become active Finallythe acquisition category indicates if the type of collected datais homogeneous or heterogeneous

Deployment The majority of available sensors are built-inand embedded in mobile devices Nevertheless non-embeddedsensing elements exist and are designed for very specificpurposes For this type of sensing equipment vendors typicallydo not have an interest in large scale production For examplethe gluten sensor comes as a standalone device and it isdesigned to work in couple with smartphones that receive storeand process food records of gluten detection [116] Thereforeit becomes necessary to distinguish between popular sensorstypically embedded in mobile devices and specific ones that aremostly standalone and connected to the device As mentionedin [284] sensors can be deployed either in dedicated or non-dedicated forms The latter definition includes sensors that areemployed for multiple purposes while dedicated sensors aretypically designed for a specific task

Non-dedicated Integrating non-dedicated sensors into mobiledevices is nowadays common practice Sensors are essentialfor ordinary operations of smartphones (eg the microphonefor phone calls) for social purposes (eg the camera to takepictures and record videos) and for user applications (egGPS for navigation systems) These sensors do not require tobe paired with other devices for data delivery but exploit thecommunication capabilities of mobile devices

Dedicated These sensors are typically standalone devicesdedicated to a specific purpose and are designed to be pairedwith a smartphone for data transmission Indeed they are verysmall devices with limited storage capabilities Communicationsrely on wireless technologies like Bluetooth or Near FieldCommunications (NFC) Nevertheless specific sensors such asthe GasMobile hardware architecture can be wired connectedwith smartphones through USBs [21] Whereas non-dedicatedsensors are used only for popular applications the design of

dedicated sensors is very specific and either single individualsor the entire community can profit To illustrate only the celiaccommunity can take advantage of the gluten sensor [116] whilean entire city can benefit from the fine dust sensors [127] ornuclear radiation monitoring [120]

Activity Nowadays mobile devices require a growing numberof basic sensors to operate that provide basic functionalitiesand cannot be switched off For example auto-adjusting thebrightness of the screen requires the ambient light sensor beingactive while understanding the orientation of the device canbe possible only having accelerometer and gyroscope workingcontinuously Conversely several other sensors can be switchedon and off manually by user intervention taking a pictureor recording a video requires the camera being active onlyfor a while We divide these sensors between always-on andon demand sensors This classification unveils a number ofproperties For example always-on sensors perform sensingcontinuously and they operate consuming a small amountof energy Having such deep understanding helps devisingapplications using sensor resources properly such as exploitingan always-on and low consuming sensor to switch on or offanother one For instance turning off the screen using theproximity sensor helps saving battery lifetime as the screen isa major cause of energy consumption [285] Turning off theambient light sensor and the GPS when the user is not movingor indoor permits additional power savings

Always-on These sensors are required to accomplish mobiledevices basic functionalities such as detection of rotationand acceleration As they run continuously and consume aminimal amount of energy it has become convenient forseveral applications to make use of these sensors in alsoin different contexts Activity recognition such as detectionof movement patterns (eg walkingrunning [286]ndash[288]) oractions (eg driving riding a car or sitting [289]ndash[291]) isa very important feature that the accelerometers enable Torecognize user activities some works also exploit readingsfrom pressure sensor [83] Furthermore if used in pair withthe gyroscope sensing readings from the accelerometers enableto monitor the user driving style [80] Also some applicationsuse these sensors for context-awareness and energy savingsdetecting user surroundings and disabling not needed sensors(eg GPS in indoor environments) [292]

On demand On demand sensors need to be switched onby users or exploiting an application running in backgroundTypically they serve more complex applications than always-on sensors and consume a higher amount of energy Henceit is better to use them only when they are needed As aconsequence they consume much more energy and for thisreason they are typically disabled for power savings Typicalrepresentative examples are the camera for taking a picturethe microphone to reveal the level of noise in dB and theGPS to sense the exact position of a mobile device whilesensing something (eg petrol prices in a gas station locatedanywhere [131])

Acquisition When organizers design a sensing campaignimplicitly consider which is the data necessary and in turns




Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic

Fig 14 Taxonomies on sensing layer which comprises elements and sampling process categories The elements-related taxonomies are divided into deploymentactivity and acquisition classes while sampling-related taxonomies are composed of frequency responsibility and user involvement classes

the required sensors This category identifies if the acquisitionprovides homogeneous data or heterogeneous Different sensorsgenerate different types of data often non-homogeneous Forinstance a microphone can be used to record an audio file orto sense the level of noise measured in dB

Homogeneous We classify as homogeneous a data acquisitionwhen it involves only one type of data and it does notchange from one user to another one For instance air qualitymonitoring [128] is a typical example of this category becauseall the users sense the same data using dedicated sensorsconnected to their mobile devices

Heterogeneous Data acquisition is heterogeneous when itinvolves different data types usually sampled from severalsensors Typical examples include all the applications that em-ploy a thread running on the background and infer informationprocessing data after the sensing

2) Sampling Process The sampling process category investi-gates the decision-making process and is broadly subdivided infrequency responsibility and user involvement The lower partof Fig 14 shows the taxonomies of this category In frequencywe consider the sampling decision which is divided betweencontinuous or event-based sensing The category responsibilityfocuses on the entity that is responsible for deciding to senseor not The class user involvement analyzes if the decision ofsampling is taken by the users actively or with an applicationrunning in the background

Frequency It indicates the frequency of sensing In thiscategory we analyze how often a task must be executed Sometypes of tasks can be triggered by event occurrence or becausethe device is in a particular context On the other hand sometasks need to be performed continuously

Continuous A continuous sensing indicates tasks that areaccomplished regularly and independently by the context of thesmartphone or the user activities As once the sensors involvedin the sensing process are activated they provide readings at a

given sampling rate The data collection continues until thereis a stop from the central collector (eg the quantity of data isenough) or from the user (eg when the battery level is low)In continuous sensing it is very important to set a samplingperiod that should be neither too low nor too high to result ina good choice for data accuracy and in the meanwhile not tooenergy consuming For instance air pollution monitoring [56]must be based on continuous sensing to have relevant resultsand should be independent of some particular eventEvent-based The frequency of sensing execution is event-basedwhen data collection starts after a certain event has occurredIn this context an event can be seen as an active action from auser or the central collector but also a given context awareness(eg users moving outdoor or getting on a bus) As a resultthe decision-making process is not regular but it requires anevent to trigger the process When users are aware of theircontext and directly perform the sensing task typical examplesare to detect food allergies [116] and to take pictures after acar accident or a natural disaster like an earthquake Whena user is not directly involved in the sensing tasks can beevent-based upon recognition of the context (eg user gettingon a bus [142])

Responsibility It defines the entity that is responsible fortaking the decision of sensing On the one hand mobile devicescan take proper decisions following a distributed paradigmOn the other hand the central collector can be designed tobe responsible for taking sensing decisions in a centralizedfashion The collector has a centralized view of the amount ofinformation already collected and therefore can distribute tasksamong users or demand for data in a more efficient mannerMobile devices Devices or users take sampling decisionslocally and independently from the central authority Thesingle individual decides actively when where how andwhat to sample eg taking a picture for sharing the costsof goods [132] When devices take sampling decisions it isoften necessary to detect the context in which smartphonesand wearable devices are The objective is to maximize the




utility of data collection and minimizing the cost of performingunnecessary operationsCentral collector In the centralized approach the collector takesdecisions about sensing and communicate them to the mobiledevices Centralized decisions can fit both participatory andopportunistic paradigms If the requests are very specific theycan be seen as tasks by all means and indeed the centralizeddecision paradigm suits very well a direct involvement of usersin sensing However if the requests are not very specific butcontain generic information a mobile application running inthe background is exploited

User involvement In MCS literature user involvement isa very generic concept that can assume different meaningsaccording to the context in which it is considered In additionparticipatory sensing is often used only to indicate that sensingis performed by participants [58] We use the term userinvolvement to define if a sensing process requires or not activeactions from the devicersquos owner We classify as opportunisticthe approach in which users are not directly involved in theprocess of gathering data and usually an application is run inbackground (eg sampling user movements from the GPS)The opposite approach is the participatory one which requiresan active action from users to gather information (eg takinga picture of a car accident) In the following we discuss bothapproaches in details giving practical examplesParticipatory approach It requires active actions from userswho are explicitly asked to perform specific tasks They areresponsible to consciously meet the application requests bydeciding when what where and how to perform sensingtasks (eg to take some pictures with the camera due toenvironmental monitoring or record audio due to noiseanalysis) Upon the sensing tasks are submitted by the MCSplatform the users take an active part in the task allocationprocess as manually decide to accept or decline an incomingsensing task request [10] In comparison to the opportunisticapproach complex operations can be supported by exploitinghuman intelligence who can solve the problem of contextawareness in a very efficient way Multimedia data can becrowdsensed in a participatory manner such as images for pricecomparison [133] or audio signals for noise analysis [123]) [58]Participation is at the usersrsquo discretion while the users are tobe rewarded on the basis of their level of participation as wellas the quality of the data they provide [11] In participatorysensing making sensing decisions is only under the userrsquosdiscretion hence context-awareness is not a critical componentas opposed to the opportunistic sensing [51] As data accuracyaims at minimum estimation error when crowdsensed data isaggregated to sense a phenomenon participatory sensing canavoid extremely low accuracies by human intervention [35] Agrand challenge in MCS occurs due to the battery limitationof mobile devices When users are recruited in a participatorymanner the monitoring and control of battery consumption arealso delegated from the MCS platform to the users who areresponsible for avoiding transmitting low-quality dataOpportunistic approach In this approach users do not havedirect involvement but only declare their interest in joininga campaign and providing their sensors as a service Upon

a simple handshake mechanism between the user and theMCS platform a MCS thread is generated on the mobiledevice (eg in the form of a mobile app) and the decisionsof what where when and how to perform the sensing aredelegated to the corresponding thread After having acceptedthe sensing process the user is totally unconscious withno tasks to perform and data collection is fully automatedEither the MCS platform itself or the background thread thatcommunicates with the MCS platform follows a decision-making procedure to meet a pre-defined objective function [37]The smartphone itself is context-aware and makes decisionsto sense and store data automatically determining when itscontext matches the requirements of an application Thereforecoupling opportunistic MCS systems with context-awarenessis a crucial requirement Hence context awareness serves as atool to run predictions before transmitting sensed data or evenbefore making sensing decisions For instance in the case ofmultimedia data (eg images video etc) it is not viable toreceive sensing services in an opportunistic manner Howeverit is possible to detect road conditions via accelerometer andGPS readings without providing explicit notification to theuser [293] As opposed to the participatory approach the MCSplatform can distribute the tasks dynamically by communicatingwith the background thread on the mobile device and byconsidering various criteria [180] [209] [210] As opportunisticsensing completely decouples the user and MCS platformas a result of the lack of user pre-screening mechanism onthe quality of certain types of data (eg images video etc)energy consumption might be higher since even low-qualitydata will be transmitted to the MCS platform even if it willbe eventually discarded [294] Moreover the thread that isresponsible for sensing tasks continues sampling and drainingthe battery In order to avoid such circumstances energy-awareMCS strategies have been proposed [295] [296]

B Classification

This section surveys literature works according to the sensinglayer taxonomy previously proposed

Elements It classifies the literature works according to the tax-onomy presented in VII-A1 The characteristics of each paperare divided between the subcategories shown in Table XIIIDeployment Non-dedicated sensors are embedded in smart-phones and typically exploited for context awareness Ac-celerometer can be used for pattern recognition in healthcareapplications [17] [136] road and traffic conditions [20] [142][145] The microphone can be useful for mapping the noisepollution in urban environments [115] [122] [123] or torecognize sound patterns [166] In some applications usingspecialized sensors requires a direct involvement of usersFor instance E-commerce exploits the combination betweencamera and GPS [131]ndash[133] while mobile social networksutilize most popular built-in sensors [60] [149] [150] [153][155] Specialized sensing campaigns employ dedicated sensorswhich are typically not embedded in mobile devices andrequire to be connected to them The most popular applicationsthat deploy dedicated sensors are environmental monitoringsuch as air quality [21] [56] nuclear radiations [120] space




TABLE XIIICLASSIFICATION BASED ON ELEMENTS TAXONOMY OF SENSING LAYER

DEPLOYMENT ACTIVITY ACQUISITION

PROJECT REFERENCE Dedicated Non-dedicated Always-on On demand Homogeneous Heterogeneous


weather [129] and emergencies management like floodings [24]Other MCS systems that exploit dedicated sensors are in healthcare [134] eg to check the quantity of gluten in food [116]for celiacs

Activity For everyday usage mobile devices leverage on always-on and low consuming sensors which are also employedby many different applications For instance accelerometeris used to detect road conditions [20] and microphone fornoise pollution [115] [121] [122] [129] Typical examplesusing on-demand sensors are mobile social networks [60][150] [153] [155] e-commerce [132] [133] and wellbeingapplications [17] [19] [136]

Acquisition Homogeneous acquisition includes MCS systemsthat collect data of one type For instance comparing pricesof goods or fuel requires only the info of the price whichcan be collected through an image or bar code [131] [132]Noise and air monitoring are other typical examples thatsample respectively a value in dB [122] [123] and measure ofair pollution [21] Movi [154] consists in contributing videocaptured from the camera through collaborative sensing Someapplications on road monitoring are also homogeneous forinstance to monitor traffic conditions [20] Other applications

exploiting only a type of data can be mapping WiFi signal [164]or magnetometer [297] for localization enhancing locationreliability [168] or sound patterns [167] Most of MCS cam-paigns require multiple sensors and a heterogeneous acquisitionTypical applications that exploit multiple sensors are mobilesocial networks [60] [150] [151] [153] [155] Health careapplications typically require interaction between differentsensors such as accelerometer to recognize movement patternsand specialized sensors [17] [19] [136] For localizationboth GPS and WiFi signals can be used [298] Intelligenttransportation systems usually require the use of multiplesensors including predicting bus arrival time [142] monitoringthe traffic condition [145]

Sampling Sampling category surveys and classifies worksaccording to the taxonomy presented in VII-A2 The charac-teristics of each paper divided between the subcategories areshown in Table XIV

Frequency A task is performed continuously when it hastemporal and spatial constraints For instance monitoringnoise [115] [122] [123] air pollution [21] road and trafficconditions [20] [145] require ideally to collect informationcontinuously to receive as much data as possible in the whole




TABLE XIVCLASSIFICATION BASED ON SAMPLING TAXONOMY OF SENSING LAYER

FREQUENCY RESPONSIBILITY USER INVOLVEMENT

PROJECT REFERENCE Continuous Event-based Local Centralized Participatory Opportunistic


region of interest and for the total sensing time to accuratelymap the phenomena under analysis Other applications requir-ing a continuous sensing consist in characterizing the coverageof WiFi intensity [164] While mobile social networks usuallyrepresent event-based sensing some exceptions presents acontinuous contribution [149] [151]ndash[153] Event-based MCSsystems sense and report data when a certain event takes placewhich can be related to context awareness for instance whenautomatically detecting a car accident [143] or to a direct actionfrom a user such as recommending travels to others [169]Typical examples are mobile social networks [59] [60] [150][155] or health care applications [17] [19] [136] where userssense and share in particular moments Other examples arecomparing the prices of goods [132] [133] monitoring busarrival time while waiting [142] and sending audio [167] orvideo [154] samples

Responsibility The responsibility of sensing and reporting is onmobile devices when users or the device itself with an applica-tion running on background decide when to sample accordingto the allocated task independently to the central collector or

other devices The decision process typically depends on mobiledevices in mobile social networks [59] [60] [150] [151] [153][155] healthcare [17] [19] [136] e-commerce [132] [133]and environmental and road monitoring [20] [21] [115] [122][123] [142] [143] The decision depends on the central collectorwhen it is needed a general knowledge of the situation forinstance when more samples are needed from a certain regionof interest Characterizing WiFi [164] space weather [129]estimating the traffic delay [145] or improving the locationreliability [168] require a central collector approach

User involvement It includes the participatory approach andthe opportunistic one MCS systems require a participatoryapproach when users exploit a sensor that needs to beactivated or when human intelligence is required to detect aparticular situation A typical application that usually requiresa participatory approach is health care eg to take picturesfor a diet [17] [19] [136] In mobile social networks it isalso required to share actively updates [60] [62] [151] [153][155] Some works require users to report their impact aboutenvironmental [121] or emotional situations [149] Comparing




prices of goods requires users taking pictures and sharingthem [133] [132] Opportunistic approach is employed whendevices are responsible for sensing Noise [115] [122] [123]and air quality [21] [22] [56] monitoring map places withautomatic sampling performed by mobile devices Intelligenttransportation systems also typically exploit an applicationrunning on the background without any user interventionsuch as monitoring traffic and road conditions [20] [145]or bus arrival time [142] Mapping WiFi intensity is anotherapplication that does not require active user participation [164]

VIII DISCUSSION

The objective of this section is twofold On the one handwe provide a retrospective analysis of the past MCS researchto expose the most prominent upcoming challenges andapplication scenarios On the other hand we present inter-disciplinary connections between MCS and other researchareas

A Looking Back to See Whatrsquos Next

A decade has passed since the appearance of the first pillarworks on MCS In such a period a number of successfuland less successful proposals were developed hence in thefollowing paragraphs we summarize main insights and lessonslearned In these years researchers have deeply investigatedseveral aspects (eg user recruitment task allocation incentivesmechanisms for participation privacy) of MCS solutions indifferent application domains (eg healthcare intelligent trans-portation systems public safety) In the meanwhile the MCSscenario has incredibly evolved because of a number of factorsFirst new communications technologies will lay the foundationof next-generation MCS systems such as 5G Device-to-Device(D2D) communications Mobile Edge Computing (MEC)Second smart mobility and related services are modifying thecitizensrsquo behavior in moving and reaching places Bike sharingcarpooling new public transport modalities are rapidly andconsistently modifying pedestrian mobility patterns and placesreachability This unleashes a higher level of pervasivenessfor MCS systems In addition more sources to aggregatedata to smart mobile devices should be taken into account(eg connected vehicles) All these considerations influenceconsiderably MCS systems and the approach of organizers todesign a crowdsensing campaign

One of the most challenging issues MCS systems face isto deal with data potentially unreliable or malicious due tocheap sensors or misleading user behavior Mobile devicesmeasurements are simply imperfect because their standardsensors are mostly not designed for scientific applicationsbut considering size cheap cost limited power consumptionand functionality To give some representative examplesaccelerometers are typically affected by basic signal processingproblems such as noise temporal jitter [118] [299] whichconsequently provide incomplete and unreliable data limitingthe accuracy of several applications eg activity recognitionroad surface monitoring and everything related to mobile patternrecognition

MOBILE

CROWDSENSING

FOG amp MOBILEMULTI-ACCESS

EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP

LEARNING ampPREDICTIVEANALYTICS SMART CITIES

ANDURBAN

COMPUTING

5G NETWORKS

Fig 15 Connections with other research areas

Despite the challenges MCS has revealed a win-win strategywhen used as a support and to improve existing monitoringinfrastructure for its capacity to increase coverage and context-awareness Citizens are seen as a connection to enhancethe relationship between a city and its infrastructure Togive some examples crowdsensed aggregated data is used incivil engineering for infrastructure management to observethe operational behavior of a bridge monitoring structuralvibrations The use case in the Harvard Bridge (Boston US)has shown that data acquired from accelerometer of mobiledevices in cars provide considerable information on the modalfrequencies of the bridge [30] Asfault is a system that monitorsroad conditions by performing ML and signal processingtechniques on data collected from accelerometers embeddedin mobile devices [300] Detection of emergency situationsthrough accelerometers such as earthquakes [117] [119] areother fields of application Creekwatch is an application thatallows monitoring the conditions of the watershed throughcrowdsensed data such as the flow rate [24] Safestreet isan application that aggregates data from several smartphonesto monitor the condition of the road surface for a saferdriving and less risk of car accidents [144] Glutensensorcollects information about healthy food and shares informationbetween celiac people providing the possibility to map and raterestaurants and places [116] CrowdMonitor is a crowdsensingapproach in monitoring the emergency services through citizensmovements and shared data coordinating real and virtualactivities and providing an overview of an emergency situationoverall in unreachable places [301]

B Inter-disciplinary Interconnections

This section analyzes inter-disciplinary research areas whereMCS plays an important role (eg smart cities and urbancomputing) or can benefit from novel technological advances(eg 5G networks and machine learning) Fig 15 providesgraphical illustration of the considered areas

Smart cities and urban computing Nowadays half of theworld population lives in cities and this percentage is projected




to rise even more in the next years [302] While nearly 2 ofthe worldrsquos surface is occupied by urban environments citiescontribute to 80 of global gas emission 75 of global energyconsumption [303] and 60 of residential water use [302] Forthis reason sustainable development usually with Informationand Communication Technologies (ICT) plays a crucial rolein city development [304] While deploying new sensinginfrastructures is typically expensive MCS systems representa win-win strategy The interaction of human intelligenceand mobility with sensing and communication capabilitiesof mobile devices provides higher accuracy and better contextawareness compared to traditional sensor networks [305] Urbancomputing has the precise objective of understanding andmanaging human activities in urban environments This involvesdifferent aspects such as urban morphology and correspondingstreet network (it has been proven that the configuration ofthe street network and the distribution of outdoor crimesare associated [162]) relation between citizens and point ofinterests (POIs) commuting required times accessibility andavailability of transportations [306] (eg public transportscarpooling bike sharing) Analyzing patterns of human flowsand contacts dynamics of residents and non-residents andcorrelations with POIs or special events are few examples ofscenarios where MCS can find applicability [307]

Big data machinedeep learning and predictive analyticsAccording to Cisco forecasts the total amount of data createdby Internet-connected devices will increase up to 847 ZB peryear by 2021 [308] Such data originates from a wide range offields and applications and exhibit high heterogeneity large vol-ume variety uncertain accuracy and dependency on applicationrequirements Big data analytics aims to inspect the contributeddata and gain insights applicable to different purposes suchas monitoring phenomena profiling user behaviors unravelinginformation that leads to better conclusions than raw data [309]In the last years big data analytics has revealed a successfulemerging strategy in smart and connected communities suchas MCS systems for real use cases [310]

Machine and deep learning techniques allow managing largeand heterogeneous data volumes and applications in complexmobile architectures [311] such as MCS systems Different MLtechniques can be used to optimize the entire cycle of MCSparadigm in particular to maximize sensing quality whileminimizing costs sustained by users [312] Deep learning dealswith large datasets by using back-propagation algorithms andallows computational models that are composed of multipleprocessing layers to learn representations of data with multiplelevels of abstraction [313] Crowdsensed data analyzed withneural networks techniques permits to predict human perceptualresponse to images [314]

Predictive analytics aims to predict future outcomes andevents by exploiting statistical modeling and deep learningtechniques For example foobot exploits learning techniques todetect patterns of air quality [315] In intelligent transportationsystems deep learning techniques can be used to analyzespatio-temporal correlations to predict traffic flows and improvetravel times [316] and to prevent pedestrian injuries anddeaths [317] In [318] the authors propose to take advantage

of sensed data spatio-temporal correlations By performingcompressed-sensing and through deep Q-learning techniquesthe system infers the values of sensing readings in uncoveredareas In [319] the authors indirectly rely on transfer learningto learns the parameters of the crowdsensing network frompreviously acquired data The objective is to model and estimatethe parameters of an unknown model In [320] the authorspropose to efficiently allocate tasks according to citizensrsquobehavior and profile They aim to profile usersrsquo preferencesexploiting implicit feedback from their historical performanceand to formalize the problem of usersrsquo reliability through asemi-supervised learning model

Distributed payment platforms based on blockchains smart contracts for data acquisition Monetary reward-ing ie to distribute micro-payments according to specificcriteria of the sensing campaign is certainly one of themost powerful incentives for recruitment To deliver micro-payments among the participants requires distributed trustedand reliable platforms that run smart contracts to move fundsbetween the parties A smart contract is a self-executing digitalcontract verified through peers without the need for a centralauthority In this context blockchain technology providesthe crowdsensing stakeholders with a transparent unalterableordered ledger enabling a decentralized and secure environmentIt makes feasible to share services and resources leading to thedeployment of a marketplace of services between devices [321]Hyperledger is an open-source project to develop and improveblockchain frameworks and platforms [322] It is based onthe idea that only collaborative software development canguarantee interoperability and transparency to adopt blockchaintechnologies in the commercial mainstream Ethereum is adecentralized platform that allows developers to distributepayments on the basis of previously chosen instructions withouta counterparty risk or the need of a middle authority [323]These platforms enable developers to define the reward thecitizens on the basis of several parameters that can be set onapplication-basis (eg amount of data QoI battery drain)

Fog computing and mobile edge computingmulti-accessedge computing (MEC) The widespread diffusion of mobileand IoT devices challenges the cloud computing paradigm tofulfill the requirements for mobility support context awarenessand low latency that are envisioned for 5G networks [324][325] Moving the intelligence closer to the mobile devicesis a win-win strategy to quickly perform MCS operationssuch as recruitment in fog platforms [277] [326] [327] Afog platform typically consists of many layers some locatedin the proximity of end users Each layer can consist of alarge number of nodes with computation communicationand storage capabilities including routers gateways accesspoints and base stations [328] While the concept of fogcomputing was introduced by Cisco and it is seen as anextension of cloud computing [329] MEC is standardized byETSI to bring application-oriented capabilities in the core ofthe mobile operatorsrsquo networks at a one-hop distance from theend-user [330] To illustrate with few representative examplesRMCS is a Robust MCS framework that integrates edgecomputing based local processing and deep learning based data




validation to reduce traffic load and transmission latency [331]In [278] the authors propose a MEC architecture for MCSsystems that decreases privacy threats and permits citizens tocontrol the flow of contributed sensor data EdgeSense is acrowdsensing framework based on edge computing architecturethat works on top of a secured peer-to-peer network over theparticipants aiming to extract environmental information inmonitored areas [332]

5G networks The objective of the fifth generation (5G) ofcellular mobile networks is to support high mobility massiveconnectivity higher data rates and lower latency Unlikeconventional mobile networks that were optimized for aspecific objective (eg voice or data) all these requirementsneed to be supported simultaneously [333] As discussed inSubsection VI-A1 significant architectural modifications to the4G architecture are foreseen in 5G While high data rates andmassive connectivity requirements are extensions of servicesalready supported in 4G systems such as enhanced mobilebroadband (eMBB) and massive machine type communications(mMTC) ultra-reliable low-latency communications (URLLC)pose to 5G networks unprecedented challenges [334] For theformer two services the use of mm-wave frequencies is ofgreat help [335] Instead URLLC pose particular challenges tothe mobile network and the 3GPP specification 38211 (Release15) supports several numerologies with a shorter transmissiontime interval (TTI) The shorter TTI duration the shorter thebuffering and the processing times However scheduling alltraffic with short TTI duration would adversely impact mobilebroadband services like crowdsensing applications whose trafficcan either be categorized as eMBB (eg video streams) ormMTC (sensor readings) Unlike URLLC such types of trafficrequire a high data rate and benefit from conventional TTIduration

IX CONCLUDING REMARKS

As of the time of this writing MCS is a well-establishedparadigm for urban sensing In this paper we provide acomprehensive survey of MCS systems with the aim toconsolidate its foundation and terminology that in literatureappear to be often obscure of used with a variety of meaningsSpecifically we overview existing surveys in MCS and closely-related research areas by highlighting the novelty of our workAn important part of our work consisted in analyzing thetime evolution of MCS during the last decade by includingpillar works and the most important technological factorsthat have contributed to the rise of MCS We proposed afour-layered architecture to characterize the works in MCSThe architecture allows to categorize all areas of the stackfrom the application layer to the sensing and physical-layerspassing through data and communication Furthermore thearchitecture allows to discriminate between domain-specificand general-purpose MCS data collection frameworks Wepresented both theoretical works such as those pertaining tothe areas of operational research and optimization and morepractical ones such as platforms simulators and datasets Wealso provided a discussion on economics and data trading Weproposed a detailed taxonomy for each layer of the architecture

classifying important works of MCS systems accordingly andproviding a further clarification on it Finally we provided aperspective of future research directions given the past effortsand discussed the inter-disciplinary interconnection of MCSwith other research areas

REFERENCES

[1] H Falaki D Lymberopoulos R Mahajan S Kandula and D EstrinldquoA first look at traffic on smartphonesrdquo in Proceedings of the 10th ACMSIGCOMM Conference on Internet Measurement ser IMC 2010 pp281ndash287

[2] H T Dinh C Lee D Niyato and P Wang ldquoA survey of mobilecloud computing architecture applications and approachesrdquo WirelessCommunications and Mobile Computing vol 13 no 18 pp 1587ndash16112013

[3] ldquoGartner says global smartphone sales stalled in thefourth quarter of 2018rdquo Gartner 2019 [Online] Avail-able httpswwwgartnercomennewsroompress-releases2019-02-21-gartner-says-global-smartphone-sales-stalled-in-the-fourth-quart

[4] ldquoGartner says worldwide wearable device sales to grow26 percent in 2019rdquo Gartner 2018 [Online] Avail-able httpswwwgartnercomennewsroompress-releases2018-11-29-gartner-says-worldwide-wearable-device-sales-to-grow-

[5] ldquoWearable technology statistics and trends 2018rdquo smartinsights 2017[Online] Available httpswwwsmartinsightscomdigital-marketing-strategywearables-statistics-2017

[6] MarketsandMarkets ldquoCrowd analytics market worth $11425 millionby 2021rdquo 2018 [Online] Available httpswwwmarketsandmarketscomPressReleasescrowd-analyticsasp

[7] R Ganti F Ye and H Lei ldquoMobile crowdsensing current state andfuture challengesrdquo IEEE Communications Magazine vol 49 no 11pp 32ndash39 Nov 2011

[8] N Lane E Miluzzo H Lu D Peebles T Choudhury and A CampbellldquoA survey of mobile phone sensingrdquo IEEE Communications Magazinevol 48 no 9 pp 140ndash150 Sep 2010

[9] W Khan Y Xiang M Aalsalem and Q Arshad ldquoMobile phonesensing systems A surveyrdquo IEEE Communications Surveys Tutorialsvol 15 no 1 pp 402ndash427 First Quarter 2013

[10] B Guo Z Yu X Zhou and D Zhang ldquoFrom participatory sensing tomobile crowd sensingrdquo in IEEE International Conference on PervasiveComputing and Communications Workshops (PERCOM Workshops)Mar 2014 pp 593ndash598

[11] J Sun ldquoIncentive mechanisms for mobile crowd sensing Currentstates and challenges of workrdquo CoRR vol abs13108364 2013[Online] Available httparxivorgabs13108364

[12] X Zhang Z Yang W Sun Y Liu S Tang K Xing and X MaoldquoIncentives for mobile crowd sensing A surveyrdquo IEEE CommunicationsSurveys Tutorials vol 18 no 1 pp 54ndash67 First Quarter 2016

[13] L Jaimes I Vergara-Laurens and A Raij ldquoA survey of incentivetechniques for mobile crowd sensingrdquo IEEE Internet of Things Journalvol 2 no 5 pp 370ndash380 Oct 2015

[14] K Ota M Dong J Gui and A Liu ldquoQuoin Incentive mechanisms forcrowd sensing networksrdquo IEEE Network vol 32 no 2 pp 114ndash119Mar 2018

[15] L Pournajaf L Xiong D A Garcia-Ulloa and V Sunderam ldquoAsurvey on privacy in mobile crowd sensing task managementrdquo TechnicalReport TR-2014-002 Department of Mathematics and Computer ScienceEmory University 2014

[16] D Christin A Reinhardt S S Kanhere and M Hollick ldquoA surveyon privacy in mobile participatory sensing applicationsrdquo Journal ofSystems and Software vol 84 no 11 pp 1928ndash1946 2011

[17] C Gao F Kong and J Tan ldquoHealthaware Tackling obesity withhealth aware smart phone systemsrdquo in IEEE International Conferenceon Robotics and Biomimetics (ROBIO) 2009 pp 1549ndash1554

[18] G Chen B Yan M Shin D Kotz and E Berke ldquoMPCS Mobile-phone based patient compliance system for chronic illness carerdquo inIEEE Annual International Mobile and Ubiquitous Systems Networkingamp Services 2009 pp 1ndash7

[19] S Reddy A Parker J Hyman J Burke D Estrin and M HansenldquoImage browsing processing and clustering for participatory sensinglessons from a DietSense prototyperdquo in Proceedings of the ACMWorkshop on Embedded Networked Sensors 2007 pp 13ndash17




[20] P Mohan V N Padmanabhan and R Ramjee ldquoNericell richmonitoring of road and traffic conditions using mobile smartphonesrdquoin Proceedings of the ACM Conference on Embedded Network SensorSystems 2008 pp 323ndash336

[21] D Hasenfratz O Saukh S Sturzenegger and L Thiele ldquoParticipatoryair pollution monitoring using smartphonesrdquo in Mobile Sensing FromSmartphones and Wearables to Big Data ACM Apr 2012

[22] V Sivaraman J Carrapetta K Hu and B G Luxan ldquoHazeWatch Aparticipatory sensor system for monitoring air pollution in Sydneyrdquo inIEEE Conference on Local Computer Networks (LCN) - WorkshopsOct 2013 pp 56ndash64

[23] L Liu W Liu Y Zheng H Ma and C Zhang ldquoThird-Eye Amobilephone-enabled crowdsensing system for air quality monitor-ingrdquo Proceedings of the ACM on Interactive Mobile Wearable andUbiquitous Technologies vol 2 no 1 pp 1ndash26 Mar 2018

[24] S Kim C Robson T Zimmerman J Pierce and E M Haber ldquoCreekWatch Pairing usefulness and usability for successful citizen sciencerdquoin Proc of the ACM Conference on Human Factors in ComputingSystems ser CHI 2011 pp 2125ndash2134

[25] S Reddy and V Samanta ldquoUrban Sensing Garbage Watchrdquo 2011UCLA Center for Embedded Networked Sensing

[26] D Bonino M T D Alizo C Pastrone and M Spirito ldquoWasteAppSmarter waste recycling for smart citizensrdquo in International Multidisci-plinary Conference on Computer and Energy Science (SpliTech) Jul2016 pp 1ndash6

[27] O Alvear C T Calafate J-C Cano and P Manzoni ldquoCrowdsensingin smart cities Overview platforms and environment sensing issuesrdquoSensors vol 18 no 2 2018

[28] H Schaffers N Komninos M Pallot B Trousse M Nilsson andA Oliveira ldquoSmart cities and the future Internet Towards cooperationframeworks for open innovationrdquo in The Future Internet ser LectureNotes in Computer Science Springer Berlin Heidelberg 2011 vol6656 pp 431ndash446

[29] A Zanella N Bui A Castellani L Vangelista and M Zorzi ldquoInternetof Things for smart citiesrdquo IEEE Internet of Things Journal vol 1no 1 pp 22ndash32 Feb 2014

[30] T J Matarazzo P Santi S N Pakzad K Carter C Ratti B MoaveniC Osgood and N Jacob ldquoCrowdsensing framework for monitoringbridge vibrations using moving smartphonesrdquo Proceedings of the IEEEvol 106 no 4 pp 577ndash593 Apr 2018

[31] K Farkas G Feher A Benczur and C Sidlo ldquoCrowdsendingbased public transport information service in smart citiesrdquo IEEECommunications Magazine vol 53 no 8 pp 158ndash165 Aug 2015

[32] M Moreno A Skarmeta and A Jara ldquoHow to intelligently make senseof real data of smart citiesrdquo in International Conference on RecentAdvances in Internet of Things (RIoT) Apr 2015 pp 1ndash6

[33] S Nawaz C Efstratiou and C Mascolo ldquoParkSense A smartphonebased sensing system for on-street parkingrdquo in Proceedings of the ACMConference on Mobile Computing and Networking ser MobiCom 2013pp 75ndash86

[34] J Cherian J Luo H Guo S S Ho and R Wisbrun ldquoParkGaugeGauging the occupancy of parking garages with crowdsensed parkingcharacteristicsrdquo in IEEE International Conference on Mobile DataManagement (MDM) vol 1 Jun 2016 pp 92ndash101

[35] B Guo Z Wang Z Yu Y Wang N Y Yen R Huang and X ZhouldquoMobile crowd sensing and computing The review of an emerginghuman-powered sensing paradigmrdquo ACM Comput Surv vol 48 no 1pp 1ndash31 Aug 2015

[36] Y Han Y Zhu and J Yu ldquoA distributed utility-maximizing algo-rithm for data collection in mobile crowd sensingrdquo in IEEE GlobalCommunications Conference (GLOBECOM) Dec 2014 pp 277ndash282

[37] H Ma D Zhao and P Yuan ldquoOpportunities in mobile crowd sensingrdquoIEEE Communications Magazine vol 52 no 8 pp 29ndash35 Aug 2014

[38] P Jayaraman C Perera D Georgakopoulos and A ZaslavskyldquoEfficient opportunistic sensing using mobile collaborative platformMOSDENrdquo in International Conference Conference on CollaborativeComputing Networking Applications and Worksharing (Collaborate-com) Oct 2013 pp 77ndash86

[39] W Gong B Zhang and C Li ldquoTask assignment in mobile crowdsens-ing Present and future directionsrdquo IEEE Network pp 1ndash8 2018

[40] B Guo Q Han H Chen L Shangguan Z Zhou and Z Yu ldquoTheemergence of visual crowdsensing Challenges and opportunitiesrdquo IEEECommunications Surveys Tutorials vol 19 no 4 pp 2526ndash2543 FourthQuarter 2017

[41] Z Xu L Mei K-K R Choo Z Lv C Hu X Luo and Y LiuldquoMobile crowd sensing of human-like intelligence using social sensorsA surveyrdquo Neurocomputing pp ndash 2017

[42] J Phuttharak and S W Loke ldquoA review of mobile crowdsourcingarchitectures and challenges Towards crowd-empowered Internet-of-Thingsrdquo IEEE Access pp 1ndash22 Dec 2018

[43] J Liu H Shen H S Narman W Chung and Z Lin ldquoA survey ofmobile crowdsensing techniques A critical component for the Internetof Thingsrdquo ACM Transactions on Cyber-Physical Systems vol 2 no 3pp 1ndash26 Jun 2018

[44] K Abualsaud T M Elfouly T Khattab E Yaacoub L S IsmailM H Ahmed and M Guizani ldquoA survey on mobile crowd-sensing andits applications in the IoT erardquo IEEE Access vol 7 pp 3855ndash38812019

[45] I F Akyildiz W Su Y Sankarasubramaniam and E Cayirci ldquoAsurvey on sensor networksrdquo IEEE Communications Magazine vol 40no 8 pp 102ndash114 Aug 2002

[46] J Yick B Mukherjee and D Ghosal ldquoWireless sensor network surveyrdquoComputer Networks vol 52 no 12 pp 2292 ndash 2330 2008

[47] P Rawat K D Singh H Chaouchi and J M Bonnin ldquoWireless sensornetworks a survey on recent developments and potential synergiesrdquoThe Journal of Supercomputing vol 68 no 1 pp 1ndash48 Apr 2014

[48] M Liu ldquoA study of mobile sensing using smartphonesrdquo InternationalJournal of Distributed Sensor Networks vol 9 no 3 p 272916 2013

[49] C Toth and G Joacutezkoacutew ldquoRemote sensing platforms and sensors Asurveyrdquo ISPRS Journal of Photogrammetry and Remote Sensing vol115 no Supplement C pp 22 ndash 36 2016

[50] V Pejovic and M Musolesi ldquoAnticipatory mobile computing A surveyof the state of the art and research challengesrdquo ACM Comput Survvol 47 no 3 pp 1ndash29 Apr 2015

[51] N Bui M Cesana S A Hosseini Q Liao I Malanchini andJ Widmer ldquoA survey of anticipatory mobile networking Context-basedclassification prediction methodologies and optimization techniquesrdquoIEEE Communications Surveys Tutorials vol 19 no 3 pp 1790ndash1821Third Quarter 2017

[52] H Gao C H Liu W Wang J Zhao Z Song X Su J Crowcroftand K K Leung ldquoA survey of incentive mechanisms for participatorysensingrdquo IEEE Communications Surveys Tutorials vol 17 no 2 pp918ndash943 Second Quarter 2015

[53] F Restuccia S K Das and J Payton ldquoIncentive mechanisms for par-ticipatory sensing Survey and research challengesrdquo ACM Transactionson Sensor Networks (TOSN) vol 12 no 2 pp 1ndash40 Apr 2016

[54] F Restuccia N Ghosh S Bhattacharjee S K Das and T MelodialdquoQuality of information in mobile crowdsensing Survey and researchchallengesrdquo ACM Transactions on Sensor Networks (TOSN) vol 13no 4 pp 1ndash43 Nov 2017

[55] A Overeem J R Robinson H Leijnse G-J Steeneveld B P Horn andR Uijlenhoet ldquoCrowdsourcing urban air temperatures from smartphonebattery temperaturesrdquo Geophysical Research Letters vol 40 no 15pp 4081ndash4085 2013

[56] P Dutta P M Aoki N Kumar A Mainwaring C Myers W Willettand A Woodruff ldquoCommon sense participatory urban sensing usinga network of handheld air quality monitorsrdquo in Proc of the ACMConference on Embedded Networked Sensor Systems 2009 pp 349ndash350

[57] J Howe ldquoThe rise of crowdsourcingrdquo Wired Jan 2006 [Online]Available httpswwwwiredcom200606crowds

[58] J A Burke D Estrin M Hansen A Parker N Ramanathan S Reddyand M B Srivastava ldquoParticipatory sensingrdquo Center for EmbeddedNetwork Sensing 2006

[59] E Miluzzo N D Lane K Fodor R Peterson H Lu M Musolesi S BEisenman X Zheng and A T Campbell ldquoSensing meets mobile socialnetworks the design implementation and evaluation of the cencemeapplicationrdquo in Proceedings of the ACM Conference on EmbeddedNetwork Sensor Systems 2008 pp 337ndash350

[60] S Gaonkar J Li R R Choudhury L Cox and A Schmidt ldquoMicro-blog sharing and querying content through mobile phones and socialparticipationrdquo in Proc of the ACM International Conference on MobileSystems Applications and Services 2008 pp 174ndash186

[61] G Cardone L Foschini P Bellavista A Corradi C Borcea M Talasilaand R Curtmola ldquoFostering participaction in smart cities a geo-socialcrowdsensing platformrdquo IEEE Communications Magazine vol 51 no 6pp 112ndash119 Jun 2013

[62] N Vastardis and K Yang ldquoMobile social networks Architecturessocial properties and key research challengesrdquo IEEE CommunicationsSurveys amp Tutorials vol 15 no 3 pp 1355ndash1371 2013

[63] P P Jayaraman C Perera D Georgakopoulos and A Zaslavsky ldquoEffi-cient opportunistic sensing using mobile collaborative platform mosdenrdquoin International Conference Conference on Collaborative Computing




Networking Applications and Worksharing (Collaboratecom) 2013 pp77ndash86

[64] G Cardone A Cirri A Corradi and L Foschini ldquoThe ParticipActmobile crowd sensing living lab The testbed for smart citiesrdquo IEEECommunications Magazine vol 52 no 10 pp 78ndash85 Oct 2014

[65] B Kantarci and H T Mouftah ldquoTrustworthy sensing for public safetyin cloud-centric Internet of Thingsrdquo IEEE Internet of Things Journalvol 1 no 4 pp 360ndash368 Aug 2014

[66] C Tanas and J Herrera-Joancomartiacute ldquoCrowdsensing simulation usingns-3rdquo Citizen in Sensor Networks Second International WorkshopCitiSens pp 47ndash58 2014 Springer International Publishing

[67] S Chessa A Corradi L Foschini and M Girolami ldquoEmpoweringmobile crowdsensing through social and ad hoc networkingrdquo IEEECommunications Magazine vol 54 no 7 pp 108ndash114 Jul 2016

[68] C Fiandrino A Capponi G Cacciatore D Kliazovich U SorgerP Bouvry B Kantarci F Granelli and S Giordano ldquoCrowdSenSima simulation platform for mobile crowdsensing in realistic urbanenvironmentsrdquo IEEE Access vol 5 pp 3490ndash3503 Feb 2017

[69] ldquoRuntasticrdquo [Online] Available httpswwwruntasticcom[70] S Consolvo D W McDonald T Toscos M Y Chen J Froehlich

B Harrison P Klasnja A LaMarca L LeGrand R Libby et alldquoActivity sensing in the wild a field trial of UbiFit gardenrdquo in Procof the ACM Conference on Human Factors in Computing Systems serSIGCHI 2008 pp 1797ndash1806

[71] J Dai X Bai Z Yang Z Shen and D Xuan ldquoPerFallD A pervasivefall detection system using mobile phonesrdquo in IEEE InternationalConference on Pervasive Computing and Communications Workshops(PERCOM Workshops) 2010 pp 292ndash297

[72] Y He Y Li and S-O Bao ldquoFall detection by built-in tri-accelerometerof smartphonerdquo in IEEE EMBS International Conference on Biomedicaland Health Informatics (BHI) 2012 pp 184ndash187

[73] J R Kwapisz G M Weiss and S A Moore ldquoActivity recognition us-ing cell phone accelerometersrdquo ACM SigKDD Explorations Newslettervol 12 no 2 pp 74ndash82 2011

[74] M A Habib M S Mohktar S B Kamaruzzaman K S Lim T MPin and F Ibrahim ldquoSmartphone-based solutions for fall detection andprevention challenges and open issuesrdquo Sensors vol 14 no 4 pp7181ndash7208 2014

[75] S Wang C Chen and J Ma ldquoAccelerometer based transportationmode recognition on mobile phonesrdquo in Asia-Pacific Conference onWearable Computing Systems (APWCS) IEEE 2010 pp 44ndash46

[76] S Reddy J Burke D Estrin M Hansen and M Srivastava ldquoDeter-mining transportation mode on mobile phonesrdquo in IEEE InternationalSymposium on Wearable Computers (ISWC) 2008 pp 25ndash28

[77] S Hemminki P Nurmi and S Tarkoma ldquoAccelerometer-basedtransportation mode detection on smartphonesrdquo in Proc of the ACMConference on Embedded Networked Sensor Systems ser SenSys 2013p 13

[78] S Reddy M Mun J Burke D Estrin M Hansen and M SrivastavaldquoUsing mobile phones to determine transportation modesrdquo ACMTransactions on Sensor Networks (TOSN) vol 6 no 2 p 13 2010

[79] Y Michalevsky D Boneh and G Nakibly ldquoGyrophone Recognizingspeech from gyroscope signalsrdquo in Proc 23rd USENIX SecuritySymposium 2014

[80] D Johnson M M Trivedi et al ldquoDriving style recognition using asmartphone as a sensor platformrdquo in IEEE International Conferenceon Intelligent Transportation Systems (ITSC) 2011 pp 1609ndash1615

[81] H Ye T Gu X Tao and J Lu ldquoB-Loc scalable floor localizationusing barometer on smartphonerdquo in IEEE International Conference onMobile Ad Hoc and Sensor Systems (MASS) 2014 pp 127ndash135

[82] S Vanini and S Giordano ldquoAdaptive context-agnostic floor transitiondetection on smart mobile devicesrdquo in IEEE International Conferenceon Pervasive Computing and Communications Workshops (PERCOMWorkshops) 2013 pp 2ndash7

[83] K Komeda M Mochizuki and N Nishiko ldquoUser activity recognitionmethod based on atmospheric pressure sensingrdquo in Proceedings of theACM International Conference on Pervasive and Ubiquitous Computing2014 pp 737ndash746

[84] P Castillejo J-F Martiacutenez L Loacutepez and G Rubio ldquoAn Internet ofThings approach for managing smart services provided by wearabledevicesrdquo International Journal of Distributed Sensor Networks vol 9no 2 p 190813 2013

[85] X Lai Q Liu X Wei W Wang G Zhou and G Han ldquoA survey ofbody sensor networksrdquo Sensors vol 13 no 5 pp 5406ndash5447 2013

[86] M Ghamari B Janko R S Sherratt W Harwin R Piechockic andC Soltanpur ldquoA survey on wireless body area networks for eHealthcaresystems in residential environmentsrdquo Sensors vol 16 no 6 2016

[87] A Filippeschi N Schmitz M Miezal G Bleser E Ruffaldi andD Stricker ldquoSurvey of motion tracking methods based on inertialsensors A focus on upper limb human motionrdquo Sensors vol 17 no 62017

[88] E Estelleacutes-Arolas and F Gonzaacutelez-Ladroacuten-De-Guevara ldquoTowards anintegrated crowdsourcing definitionrdquo Journal of Information sciencevol 38 no 2 pp 189ndash200 Apr 2012

[89] D C Brabham ldquoCrowdsourcing as a model for problem solving Anintroduction and casesrdquo pp 75ndash90 2008

[90] J M Leimeister M Huber U Bretschneider and H Krcmar ldquoLever-aging crowdsourcing Activation-supporting components for IT-basedideas competitionrdquo Journal of Management Information Systems vol 26no 1 pp 197ndash224 2009

[91] A Kittur E H Chi and B Suh ldquoCrowdsourcing user studies withMechanical Turkrdquo in Proceedings of the ACM Conference on HumanFactors in Computing Systems ser SIGCHI 2008 pp 453ndash456

[92] J Ross L Irani M S Silberman A Zaldivar and B Tomlinson ldquoWhoare the crowdworkers Shifting demographics in mechanical turkrdquo inProceedings of the ACM Conference on Human Factors in ComputingSystems ser SIGCHI EA 2010 pp 2863ndash2872

[93] L Duan L Huang C Langbort A Pozdnukhov J Walrand andL Zhang ldquoHuman-in-the-Loop mobile networks A survey of recentadvancementsrdquo IEEE Journal on Selected Areas in Communicationsvol 35 no 4 pp 813ndash831 Apr 2017

[94] E Kamar S Hacker and E Horvitz ldquoCombining human and machineintelligence in large-scale crowdsourcingrdquo in International Conferenceon Autonomous Agents and Multiagent Systems ser AAMAS vol 1International Foundation for Autonomous Agents and MultiagentSystems 2012 pp 467ndash474

[95] R Fakoor M Raj A Nazi M Di Francesco and S K DasldquoAn integrated cloud-based framework for mobile phone sensingrdquo inProceedings of the ACM Workshop on Mobile Cloud Computing 2012pp 47ndash52

[96] D Huang T Xing and H Wu ldquoMobile cloud computing servicemodels a user-centric approachrdquo IEEE Network vol 27 no 5 pp6ndash11 Sep 2013

[97] B Kantarci and H T Mouftah ldquoSensing services in cloud-centricInternet of Things A survey taxonomy and challengesrdquo in IEEEInternational Conference on Communication Workshop (ICCW) Jun2015 pp 1865ndash1870

[98] X Sheng J Tang X Xiao and G Xue ldquoSensing as a ServiceChallenges solutions and future directionsrdquo IEEE Sensors Journalvol 13 no 10 pp 3733ndash3741 2013

[99] R D Lauro F Lucarelli and R Montella ldquoSIaaS - sensing instrumentas a service using cloud computing to turn physical instrument intoubiquitous servicerdquo in IEEE 10th International Symposium on Paralleland Distributed Processing with Applications Jul 2012 pp 861ndash862

[100] C Doukas and I Maglogiannis ldquoBringing IoT and cloud computingtowards pervasive Healthcarerdquo in International Conference on Innova-tive Mobile and Internet Services in Ubiquitous Computing Jul 2012pp 922ndash926

[101] M Boukhechba A R Daros K Fua P I Chow B A Teachman andL E Barnes ldquoDemonicSalmon Monitoring mental health and socialinteractions of college students using smartphonesrdquo in IEEEACM Con-ference on Connected Health Applications Systems and EngineeringTechnologies (CHASE) Sep 2018

[102] I Khan F Belqasmi R Glitho N Crespi M Morrow and P PolakosldquoWireless sensor network virtualization A surveyrdquo IEEE Communi-cations Surveys Tutorials vol 18 no 1 pp 553ndash576 First Quarter2016

[103] H I Kobo A M Abu-Mahfouz and G P Hancke ldquoA survey onsoftware-defined wireless sensor networks Challenges and designrequirementsrdquo IEEE Access vol 5 pp 1872ndash1899 2017

[104] I Ali A Gani I Ahmedy I Yaqoob S Khan and M H Anisi ldquoDatacollection in smart communities using sensor cloud Recent advancestaxonomy and future research directionsrdquo IEEE CommunicationsMagazine vol 56 no 7 pp 192ndash197 2018

[105] A B Zaslavsky C Perera and D Georgakopoulos ldquoSensing as aService and Big Datardquo CoRR vol abs13010159 2013 [Online]Available httparxivorgabs13010159

[106] S Alam M M R Chowdhury and J Noll ldquoSenaaS An event-driven sensor virtualization approach for Internet of Things cloudrdquoin IEEE International Conference on Networked Embedded Systems forEnterprise Applications Nov 2010 pp 1ndash6

[107] S Distefano G Merlino and A Puliafito ldquoSensing and actuationas a service A new development for cloudsrdquo in IEEE International




Symposium on Network Computing and Applications Aug 2012 pp272ndash275

[108] G Merlino S Arkoulis S Distefano C Papagianni A Puliafito andS Papavassiliou ldquoMobile crowdsensing as a service A platform forapplications on top of sensing cloudsrdquo Future Generation ComputerSystems vol 56 pp 623 ndash 639 2016

[109] ldquoMaaS as a conceptrdquo MaaS Finland [Online] Available httpmaasglobalmaas-as-a-concept

[110] S A Rahman A Mourad M E Barachi and W A Orabi ldquoA novel on-demand vehicular sensing framework for traffic condition monitoringrdquoVehicular Communications vol 12 pp 165 ndash 178 2018

[111] A Capponi C Fiandrino C Franck U Sorger D Kliazovichand P Bouvry ldquoAssessing performance of Internet of Things-basedmobile crowdsensing systems for sensing as a service applications insmart citiesrdquo in IEEE International Conference on Cloud ComputingTechnology and Science (CloudCom) Dec 2016 pp 456ndash459

[112] K Han C Zhang and J Luo ldquoTaming the uncertainty Budget limitedrobust crowdsensing through online learningrdquo IEEEACM Transactionson Networking vol 24 no 3 pp 1462ndash1475 Jun 2016

[113] S Satpathy B Sahoo and A K Turuk ldquoSensing and actuation as aservice delivery model in cloud edge centric Internet of Thingsrdquo FutureGeneration Computer Systems vol 86 pp 281 ndash 296 2018

[114] R Fakoor M Raj A Nazi M Di Francesco and S K Das ldquoAnintegrated cloud-based framework for mobile phone sensingrdquo in Procof the ACM Workshop on Mobile Cloud Computing ser MCC 2012pp 47ndash52

[115] I Schweizer R Baumlrtl A Schulz F Probst and M MuumlhlaumluserldquoNoisemap - real-time participatory noise mapsrdquo in InternationalWorkshop on Sensing Applications on Mobile Phones 2011 pp 1ndash5

[116] 6sensorlabs (2015) Gluten sensor [Online][117] J Reilly S Dashti M Ervasti J D Bray S D Glaser and A M Bayen

ldquoMobile phones as seismologic sensors Automating data extraction forthe ishake systemrdquo IEEE Transactions on Automation Science andEngineering vol 10 no 2 pp 242ndash251 Apr 2013

[118] S Dashti J D Bray J Reilly S Glaser A Bayen and E MarildquoEvaluating the reliability of phones as seismic monitoring instrumentsrdquoEarthquake Spectra vol 30 no 2 pp 721ndash742 2014

[119] M Faulkner R Clayton T Heaton K M Chandy M Kohler J BunnR Guy A Liu M Olson M Cheng and A Krause ldquoCommunity senseand response systems Your phone as quake detectorrdquo ACM Communvol 57 no 7 pp 66ndash75 Jul 2014

[120] Y Ishigaki Y Matsumoto R Ichimiya and K Tanaka ldquoDevelopmentof mobile radiation monitoring system utilizing smartphone and itsfield tests in fukushimardquo IEEE Sensors Journal vol 13 no 10 pp3520ndash3526 2013

[121] M Mun S Reddy K Shilton N Yau J Burke D Estrin M HansenE Howard R West and P Boda ldquoPEIR the personal environmentalimpact report as a platform for participatory sensing systems researchrdquoin Proc of the ACM International Conference on Mobile SystemsApplications and Services 2009 pp 55ndash68

[122] R K Rana C T Chou S S Kanhere N Bulusu and W Hu ldquoEar-phone an end-to-end participatory urban noise mapping systemrdquo in Procof the ACMIEEE International Conference on Information Processingin Sensor Networks 2010 pp 105ndash116

[123] N Maisonneuve M Stevens M E Niessen and L Steels ldquoNoiseTubeMeasuring and mapping noise pollution with mobile phonesrdquo inInformation Technologies in Environmental Engineering Springer2009 pp 215ndash228

[124] M Stevens and E DrsquoHondt ldquoCrowdsourcing of pollution data usingsmartphonesrdquo in Workshop on Ubiquitous Crowdsourcing 2010

[125] F Montori L Bedogni and L Bononi ldquoA collaborative Internet ofThings architecture for smart cities and environmental monitoringrdquoIEEE Internet of Things Journal vol 5 no 2 pp 592ndash605 Apr 2018

[126] C Xiang P Yang and S Xiao ldquoCounter-strike accurate and robustidentification of low-level radiation sources with crowd-sensing net-worksrdquo Personal and Ubiquitous Computing vol 21 no 1 pp 75ndash84Feb 2017

[127] M Budde P Barbera R El Masri T Riedel and M Beigl ldquoRetrofittingsmartphones to be used as particulate matter dosimetersrdquo in Proc ofthe ACM International Symposium on Wearable Computers ser ISWC2013 pp 139ndash140

[128] D A Galvan A Komjathy M P Hickey P Stephens J SnivelyY Tony Song M D Butala and A J Mannucci ldquoIonospheric signaturesof tohoku-oki tsunami of march 11 2011 Model comparisons near theepicenterrdquo Radio Science vol 47 no 4 2012

[129] V Pankratius F Lind A Coster P Erickson and J Semeter ldquoMobilecrowd sensing in space weather monitoring the Mahali projectrdquo IEEECommunications Magazine vol 52 no 8 pp 22ndash28 2014

[130] N Bulusu C T Chou S Kanhere Y Dong S Sehgal D Sullivan andL Blazeski ldquoParticipatory sensing in commerce Using mobile cameraphones to track market price dispersionrdquo in Proc of the InternationalWorkshop on Urban Community and Social Applications of NetworkedSensing Systems (UrbanSense) 2008 pp 6ndash10

[131] Y F Dong S Kanhere C T Chou and N Bulusu ldquoAutomaticcollection of fuel prices from a network of mobile camerasrdquo inDistributed computing in sensor systems Springer 2008 pp 140ndash156

[132] S Sehgal S S Kanhere and C T Chou ldquoMobishop Using mobilephones for sharing consumer pricing informationrdquo in Demo Sessionof the Intl Conference on Distributed Computing in Sensor Systems2008

[133] L Deng and L P Cox ldquoLivecompare grocery bargain hunting throughparticipatory sensingrdquo in Proc of the ACM Workshop on MobileComputing Systems and Applications 2009 p 4

[134] K Sha G Zhan W Shi M Lumley C Wiholm and B ArnetzldquoSPA a smart phone assisted chronic illness self-management systemwith participatory sensingrdquo in Proceedings of the ACM Workshop onSystems and Networking Support for Health Care and Assisted LivingEnvironments 2008 p 5

[135] W-J Yi W Jia and J Saniie ldquoMobile sensor data collector usingAndroid smartphonerdquo in IEEE 55th International Midwest Symposiumon Circuits and Systems (MWSCAS) 2012 pp 956ndash959

[136] J Hicks N Ramanathan D Kim M Monibi J Selsky M Hansenand D Estrin ldquoAndWellness an open mobile system for activity andexperience samplingrdquo in Wireless Health ACM 2010 pp 34ndash43

[137] Q Xu and R Zheng ldquoMobiBee A mobile treasure hunt game forlocation-dependent fingerprint collectionrdquo in Proceedings of the ACMInternational Joint Conference on Pervasive and Ubiquitous ComputingAdjunct 2016 pp 1472ndash1477

[138] B Zhou Q Li Q Mao and W Tu ldquoA robust crowdsourcing-basedindoor localization systemrdquo Sensors vol 17 no 4 p 864 2017

[139] Q Xu R Zheng and E Tahoun ldquoDetecting location fraud in indoormobile crowdsensingrdquo in Proceedings of the First ACM Workshop onMobile Crowdsensing Systems and Applications 2017 pp 44ndash49

[140] F Calabrese M Colonna P Lovisolo D Parata and C Ratti ldquoReal-time urban monitoring using cell phones A case study in Romerdquo IEEETransactions on Intelligent Transportation Systems vol 12 no 1 pp141ndash151 Mar 2011

[141] R Tse Y Xiao G Pau M Roccetti S Fdida and G Marfia ldquoOn thefeasibility of social network-based pollution sensing in ITSsrdquo arXivpreprint arXiv14116573 2014

[142] P Zhou Y Zheng and M Li ldquoHow long to wait predicting bus arrivaltime with mobile phone based participatory sensingrdquo in Proc of theACM International Conference on Mobile Systems Applications andServices 2012 pp 379ndash392

[143] J White C Thompson H Turner B Dougherty and D C SchmidtldquoWreckwatch Automatic traffic accident detection and notification withsmartphonesrdquo Mobile Networks and Applications vol 16 no 3 pp285ndash303 2011

[144] V Singh D Chander U Chhaparia and B Raman ldquoSafeStreetAn automated road anomaly detection and early-warning systemusing mobile crowdsensingrdquo in IEEE International Conference onCommunication Systems Networks (COMSNETS) Jan 2018 pp 549ndash552

[145] A Thiagarajan L Ravindranath K LaCurts S Madden H Balakrish-nan S Toledo and J Eriksson ldquoVTrack accurate energy-aware roadtraffic delay estimation using mobile phonesrdquo in Proceedings of theACM Conference on Embedded Networked Sensor Systems ser SenSys2009 pp 85ndash98

[146] M A Rahman and M S Hossain ldquoA location-based mobile crowd-sensing framework supporting a massive Ad Hoc social networkenvironmentrdquo IEEE Communications Magazine vol 55 no 3 pp76ndash85 Mar 2017

[147] Y Chon N D Lane Y Kim F Zhao and H Cha ldquoUnderstandingthe coverage and scalability of place-centric crowdsensingrdquo in Proc ofthe ACM international joint Conference on Pervasive and UbiquitousComputing ser UbiComp 2013 pp 3ndash12

[148] Y Chon N D Lane F Li H Cha and F Zhao ldquoAutomatically char-acterizing places with opportunistic crowdsensing using smartphonesrdquoin Proceedings of the ACM Conference on Ubiquitous Computing 2012pp 481ndash490

[149] K K Rachuri M Musolesi C Mascolo P J Rentfrow C Longworthand A Aucinas ldquoEmotionSense a mobile phones based adaptive




platform for experimental social psychology researchrdquo in Proceedingsof the ACM International Conference on Ubiquitous Computing 2010pp 281ndash290

[150] A-K Pietilaumlinen E Oliver J LeBrun G Varghese and C DiotldquoMobiClique middleware for mobile social networkingrdquo in Proc of theACM Workshop on Online Social Networks 2009 pp 49ndash54

[151] N Eagle and A Pentland ldquoSocial serendipity Mobilizing socialsoftwarerdquo IEEE Pervasive Computing vol 4 no 2 pp 28ndash34 2005

[152] K K Rachuri C Mascolo M Musolesi and P J Rentfrow ldquoSociable-sense exploring the trade-offs of adaptive sampling and computationoffloading for social sensingrdquo in Proceedings of the ACM InternationalConference on Mobile Computing and Networking 2011 pp 73ndash84

[153] A Beach M Gartrell S Akkala J Elston J Kelley K NishimotoB Ray S Razgulin K Sundaresan B Surendar et al ldquoWhozthatevolving an ecosystem for context-aware mobile social networksrdquo IEEENetwork vol 22 no 4 pp 50ndash55 2008

[154] X Bao and R Roy Choudhury ldquoMoVi mobile phone based videohighlights via collaborative sensingrdquo in Proceedings of the ACMInternational Conference on Mobile Systems Applications and Services2010 pp 357ndash370

[155] E Miluzzo C T Cornelius A Ramaswamy T Choudhury Z Liu andA T Campbell ldquoDarwin phones the evolution of sensing and inferenceon mobile phonesrdquo in Proc of the ACM International Conference onMobile Systems Applications and Services 2010 pp 5ndash20

[156] J Ballesteros B Carbunar M Rahman N Rishe and S IyengarldquoTowards safe cities A mobile and social networking approachrdquo IEEETransactions on Parallel and Distributed Systems vol 25 no 9 pp2451ndash2462 2014

[157] P Fraga-Lamas T M Fernaacutendez-Carameacutes M Suaacuterez-AlbelaL Castedo and M Gonzaacutelez-Loacutepez ldquoA review on Internet of Thingsfor defense and public safetyrdquo Sensors vol 16 no 10 2016

[158] B Zhang C H Liu J Tang Z Xu J Ma and W Wang ldquoLearning-based energy-efficient data collection by unmanned vehicles in smartcitiesrdquo IEEE Transactions on Industrial Informatics vol 14 no 4 pp1666ndash1676 Apr 2018

[159] Z Zhou J Feng B Gu B Ai S Mumtaz J Rodriguez andM Guizani ldquoWhen mobile crowd sensing meets UAV Energy-efficient task assignment and route planningrdquo IEEE Transactions onCommunications vol 66 no 11 pp 5526ndash5538 Nov 2018

[160] E Barka C A Kerrache N Lagraa and A Lakas ldquoBehavior-aware UAV-assisted crowd sensing technique for urban vehicularenvironmentsrdquo in 15th IEEE Annual Consumer CommunicationsNetworking Conference (CCNC) Jan 2018 pp 1ndash7

[161] Y Zheng L Capra O Wolfson and H Yang ldquoUrban computingConcepts methodologies and applicationsrdquo ACM Transactions onIntelligent Systems and Technology vol 5 no 3 pp 1ndash55 Sep 2014

[162] L Summers and S D Johnson ldquoDoes the configuration of the streetnetwork influence where outdoor serious violence takes place usingspace syntax to test crime pattern theoryrdquo Journal of QuantitativeCriminology vol 33 no 2 pp 397ndash420 Jun 2017

[163] W Guo and S Wang ldquoMobile crowd-sensing wireless activity withmeasured interference powerrdquo IEEE Wireless Communications Lettersvol 2 no 5 pp 539ndash542 2013

[164] A Farshad M K Marina and F Garcia ldquoUrban WiFi characteri-zation via mobile crowdsensingrdquo in IEEE Network Operations andManagement Symposium (NOMS) 2014 pp 1ndash9

[165] S Rosen S-J Lee J Lee P Congdon Z M Mao and K BurdenldquoMCNet Crowdsourcing wireless performance measurements throughthe eyes of mobile devicesrdquo IEEE Communications Magazine vol 52no 10 pp 86ndash91 2014

[166] H Lu W Pan N D Lane T Choudhury and A T CampbellldquoSoundSense scalable sound sensing for people-centric applications onmobile phonesrdquo in Proceedings of the ACM International Conferenceon Mobile Systems Applications and Services 2009 pp 165ndash178

[167] V Subbaraju A Kumar V Nandakumar S Batra S Kanhere P DeV Naik D Chakraborty and A Misra ldquoConferenceSense monitoringof public events using phone sensorsrdquo in Proc of the ACM Conferenceon Pervasive and Ubiquitous Computing 2013 pp 1167ndash1174

[168] M Talasila R Curtmola and C Borcea ldquoImproving location reliabilityin crowd sensed data with minimal effortsrdquo in IFIP Wireless and MobileNetworking Conference (WMNC) 2013 pp 1ndash8

[169] Z Yu Y Feng H Xu and X Zhou ldquoRecommending travel packagesbased on mobile crowdsourced datardquo IEEE Communications Magazinevol 52 no 8 pp 56ndash62 2014

[170] H Habibzadeh T Soyata B Kantarci A Boukerche and C KaptanldquoSensing communication and security planes A new challenge for a

smart city system designrdquo Computer Networks vol 144 pp 163 ndash 2002018

[171] T Anagnostopoulos A Zaslavsky K Kolomvatsos A MedvedevP Amirian J Morley and S Hadjieftymiades ldquoChallenges andopportunities of waste management in IoT-enabled smart cities Asurveyrdquo IEEE Transactions on Sustainable Computing vol 2 no 3pp 275ndash289 Jul 2017

[172] P P Jayaraman A Sinha W Sherchan S Krishnaswamy A ZaslavskyP D Haghighi S Loke and M T Do ldquoHere-n-now A framework forcontext-aware mobile crowdsensingrdquo in Proceedings of the InternationalConference on Pervasive Computing 2012

[173] J Chon and H Cha ldquoLifemap A smartphone-based context providerfor location-based servicesrdquo IEEE Pervasive Computing no 2 pp58ndash67 2011

[174] N D Lane Y Chon L Zhou Y Zhang F Li D Kim G DingF Zhao and H Cha ldquoPiggyback crowdsensing (PCS) energy efficientcrowdsourcing of mobile sensor data by exploiting smartphone appopportunitiesrdquo in Proc of the ACM Conference on Embedded NetworkedSensor Systems ser SenSys 2013

[175] A Capponi C Fiandrino D Kliazovich P Bouvry and S GiordanoldquoA cost-effective distributed framework for data collection in cloud-basedmobile crowd sensing architecturesrdquo IEEE Transactions on SustainableComputing vol 2 no 1 pp 3ndash16 Jan 2017

[176] F Montori L Bedogni and L Bononi ldquoDistributed data collectioncontrol in opportunistic mobile crowdsensingrdquo in Proc of the ACMWorkshop on Experiences with the Design and Implementation of SmartObjects ser SMARTOBJECTS 2017 pp 19ndash24

[177] H Xiong D Zhang L Wang J P Gibson and J Zhu ldquoEEMCEnabling energy-efficient mobile crowdsensing with anonymousparticipantsrdquo ACM Transactions on Intelligent Systems and Technology(TIST) vol 6 no 3 pp 1ndash26 Apr 2015 [Online] Availablehttpdoiacmorgproxybnllu1011452644827

[178] J Wang Y Wang D Zhang and S Helal ldquoEnergy saving techniquesin mobile crowd sensing Current state and future opportunitiesrdquo IEEECommunications Magazine vol 56 no 5 pp 164ndash169 May 2018

[179] C Wietfeld C Ide and B Dusza ldquoResource efficient mobile commu-nications for crowd-sensingrdquo in ACMEDACIEEE Design AutomationConference (DAC) 2014 pp 1ndash6

[180] A Chatterjee M Borokhovich L R Varshney and S VishwanathldquoEfficient and flexible crowdsourcing of specialized tasks with prece-dence constraintsrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2016 pp 1ndash9

[181] F Montori L Bedogni A D Chiappari and L Bononi ldquoSenSquareA mobile crowdsensing architecture for smart citiesrdquo in 2016 IEEE 3rdWorld Forum on Internet of Things (WF-IoT) Dec 2016 pp 536ndash541

[182] A Zaslavsky P P Jayaraman and S Krishnaswamy ldquoSharelikescrowdMobile analytics for participatory sensing and crowd-sourcing ap-plicationsrdquo in IEEE International Conference on Data EngineeringWorkshops (ICDEW) 2013 pp 128ndash135

[183] Y Xiao P Simoens P Pillai K Ha and M Satyanarayanan ldquoLoweringthe barriers to large-scale mobile crowdsensingrdquo in Proc of the ACMWorkshop on Mobile Computing Systems and Applications 2013 p 9

[184] Q Zhu M Y S Uddin N Venkatasubramanian and C HsuldquoSpatiotemporal scheduling for crowd augmented urban sensingrdquo inIEEE Conference on Computer Communications (INFOCOM) Apr2018 pp 1997ndash2005

[185] A Khan S K A Imon and S K Das ldquoEnsuring energy efficient cov-erage for participatory sensing in urban streetsrdquo in IEEE InternationalConference on Smart Computing (SMARTCOMP) 2014 pp 167ndash174

[186] V I Nistorica C Chilipirea and C Dobre ldquoHow many people areneeded for a crowdsensing campaignrdquo in IEEE 12th InternationalConference on Intelligent Computer Communication and Processing(ICCP) Sep 2016 pp 353ndash358

[187] L Wang D Zhang Y Wang C Chen X Han and A MrsquohamedldquoSparse mobile crowdsensing challenges and opportunitiesrdquo IEEECommunications Magazine vol 54 no 7 pp 161ndash167 July 2016

[188] X Tang C Wang X Yuan and Q Wang ldquoNon-interactive privacy-preserving truth discovery in crowd sensing applicationsrdquo in IEEEConference on Computer Communications (INFOCOM) Apr 2018

[189] L Cheng J Niu L Kong C Luo Y Gu W He and S K DasldquoCompressive sensing based data quality improvement for crowd-sensingapplicationsrdquo Journal of Network and Computer Applications vol 77pp 123 ndash 134 2017

[190] M Xiao J Wu S Zhang and J Yu ldquoSecret-sharing-based secure userrecruitment protocol for mobile crowdsensingrdquo in IEEE Conference onComputer Communications (INFOCOM) May 2017 pp 1ndash9




[191] J Lin M Li D Yang and G Xue ldquoSybil-proof online incentivemechanisms for crowdsensingrdquo in IEEE Conference on ComputerCommunications (INFOCOM) Apr 2018

[192] D Wu S Si S Wu and R Wang ldquoDynamic trust relationships awaredata privacy protection in mobile crowd-sensingrdquo IEEE Internet ofThings Journal vol 5 no 4 pp 2958ndash2970 Aug 2018

[193] S Bhattacharjee N Ghosh V K Shah and S K Das ldquoQnQ Qualityand quantity based unified approach for secure and trustworthy mobilecrowdsensingrdquo IEEE Transactions on Mobile Computing Dec 2018

[194] A Mehrotra S R Muumlller G M Harari S D Gosling C MascoloM Musolesi and P J Rentfrow ldquoUnderstanding the role of placesand activities on mobile phone interaction and usage patternsrdquo Pro-ceedings of the ACM on Interactive Mobile Wearable and UbiquitousTechnologies vol 1 no 3 p 84 2017

[195] W Wu J Wang M Li K Liu F Shan and J Luo ldquoEnergy-efficienttransmission with data sharing in participatory sensing systemsrdquo IEEEJournal on Selected Areas in Communications vol 34 no 12 pp4048ndash4062 Dec 2016

[196] J Wang J Tang G Xue and D Yang ldquoTowards energy-efficient taskscheduling on smartphones in mobile crowd sensing systemsrdquo ComputerNetworks vol 115 no Supplement C pp 100 ndash 109 2017

[197] Q Zhao Y Zhu H Zhu J Cao G Xue and B Li ldquoFairenergy-efficient sensing task allocation in participatory sensing withsmartphonesrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2014 pp 1366ndash1374

[198] Z Zheng F Wu X Gao H Zhu S Tang and G Chen ldquoA budgetfeasible incentive mechanism for weighted coverage maximization inmobile crowdsensingrdquo IEEE Transactions on Mobile Computing vol 16no 9 pp 2392ndash2407 Sep 2017

[199] A Obinikpo Y Zhang H Song T H Luan and B Kantarci ldquoQueuingalgorithm for effective target coverage in mobile crowd sensingrdquo IEEEInternet of Things Journal vol 4 no 4 pp 1046ndash1055 Aug 2017

[200] S He D H Shin J Zhang and J Chen ldquoNear-optimal allocationalgorithms for location-dependent tasks in crowdsensingrdquo IEEE Trans-actions on Vehicular Technology vol 66 no 4 pp 3392ndash3405 Apr2017

[201] A Ahmed K Yasumoto Y Yamauchi and M Ito ldquoDistance and timebased node selection for probabilistic coverage in people-centric sensingrdquoin IEEE Conference on Sensor Mesh and Ad Hoc Communications andNetworks Jun 2011 pp 134ndash142

[202] M Xiao J Wu H Huang L Huang and C Hu ldquoDeadline-sensitive user recruitment for mobile crowdsensing with probabilisticcollaborationrdquo in IEEE International Conference on Network Protocols(ICNP) Nov 2016 pp 1ndash10

[203] K Han C Zhang J Luo M Hu and B Veeravalli ldquoTruthful schedulingmechanisms for powering mobile crowdsensingrdquo IEEE Transactionson Computers vol 65 no 1 pp 294ndash307 Jan 2016

[204] B Liu C Song M Liu and N Liu ldquoDistinguishing uncertainobjects with multiple features for crowdsensingrdquo in IEEE GlobalCommunications Conference (GLOBECOM) 2014 pp 2751ndash2756

[205] T Zhou B Xiao Z Cai M Xu and X Liu ldquoFrom uncertain photosto certain coverage A novel photo selection approach to mobilecrowdsensingrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2018

[206] J Li Y Zhu and J Yu ldquoLoad balance vs utility maximizationin mobile crowd sensing A distributed approachrdquo in IEEE GlobalCommunications Conference (GLOBECOM) 2014 pp 265ndash270

[207] L Wang Z Yu B Guo F Yi and F Xiong ldquoMobile crowd sensing taskoptimal allocation a mobility pattern matching perspectiverdquo Frontiersof Computer Science vol 12 no 2 pp 231ndash244 Apr 2018

[208] H Xiong D Zhang G Chen L Wang V Gauthier and L E BarnesldquoiCrowd Near-optimal task allocation for piggyback crowdsensingrdquoIEEE Transactions on Mobile Computing vol 15 no 8 pp 2010ndash2022 Aug 2016

[209] Y Liu B Guo Y Wang W Wu Z Yu and D Zhang ldquoTaskme Multi-task allocation in mobile crowd sensingrdquo in Proceedings of the ACMInternational Joint Conference on Pervasive and Ubiquitous Computingser UbiComp 2016 pp 403ndash414

[210] M Xiao J Wu L Huang Y Wang and C Liu ldquoMulti-task assignmentfor crowdsensing in mobile social networksrdquo in IEEE Conference onComputer Communications (INFOCOM) Apr 2015 pp 2227ndash2235

[211] W Sun Y Zhu L M Ni and B Li ldquoCrowdsourcing sensing workloadsof heterogeneous tasks A distributed fairness-aware approachrdquo in IEEEInternational Conference on Parallel Processing Sep 2015 pp 580ndash589

[212] Q Zhao Y Zhu H Zhu J Cao G Xue and B Li ldquoFairenergy-efficient sensing task allocation in participatory sensing with

smartphonesrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2014 pp 1366ndash1374

[213] L Pu X Chen J Xu and X Fu ldquoCrowdlet Optimal worker recruitmentfor self-organized mobile crowdsourcingrdquo in IEEE Conference onComputer Communications (INFOCOM) Apr 2016 pp 1ndash9

[214] L Wang Z Yu D Zhang B Guo and C H Liu ldquoHeterogeneousmulti-task assignment in mobile crowdsensing using spatiotemporalcorrelationrdquo IEEE Transactions on Mobile Computing 2018

[215] X Wang R Jia X Tian and X Gan ldquoDynamic task assignment incrowdsensing with location awareness and location diversityrdquo in IEEEConference on Computer Communications (INFOCOM) Apr 2018

[216] J Wang L Wang Y Wang D Zhang and L Kong ldquoTask allocation inmobile crowd sensing State-of-the-art and future opportunitiesrdquo IEEEInternet of Things Journal vol 5 no 5 pp 3747ndash3757 Oct 2018

[217] R F E Khatib N Zorba and H S Hassanein ldquoCost-efficient multi-tasking in coverage-aware mobile crowd sensingrdquo in InternationalWireless Communications Mobile Computing Conference (IWCMC)Jun 2018 pp 594ndash599

[218] H Li T Li and Y Wang ldquoDynamic participant recruitment ofmobile crowd sensing for heterogeneous sensing tasksrdquo in IEEE 12thInternational Conference on Mobile Ad Hoc and Sensor Systems Oct2015 pp 136ndash144

[219] F Zhang B Jin H Liu Y W Leung and X Chu ldquoMinimum-costrecruitment of mobile crowdsensing in cellular networksrdquo in IEEEGlobal Communications Conference (GLOBECOM) Dec 2016 pp1ndash7

[220] M Karaliopoulos O Telelis and I Koutsopoulos ldquoUser recruitment formobile crowdsensing over opportunistic networksrdquo in IEEE Conferenceon Computer Communications (INFOCOM) Apr 2015 pp 2254ndash2262

[221] Z Feng Y Zhu Q Zhang L M Ni and A V Vasilakos ldquoTRACTruthful auction for location-aware collaborative sensing in mobilecrowdsourcingrdquo in IEEE International Conference on Computer Com-munications (INFOCOM) 2014 pp 1231ndash1239

[222] D Zhang H Xiong L Wang and G Chen ldquoCrowdRecruiter Selectingparticipants for piggyback crowdsensing under probabilistic coverageconstraintrdquo in ACM International Joint Conference on Pervasive andUbiquitous Computing ser UbiComp 2014 pp 703ndash714

[223] H Xiong D Zhang G Chen L Wang and V Gauthier ldquoCrowdTaskerMaximizing coverage quality in piggyback crowdsensing under budgetconstraintrdquo in IEEE International Conference on Pervasive Computingand Communications (PerCom) Mar 2015 pp 55ndash62

[224] D Yang G Xue X Fang and J Tang ldquoIncentive mechanismsfor crowdsensing Crowdsourcing with smartphonesrdquo IEEEACMTransactions on Networking vol 24 no 3 pp 1732ndash1744 Jun 2016

[225] B Guo Y Liu W Wu Z Yu and Q Han ldquoActivecrowd A frameworkfor optimized multitask allocation in mobile crowdsensing systemsrdquoIEEE Transactions on Human-Machine Systems vol 47 no 3 pp392ndash403 Jun 2017

[226] M Karaliopoulos I Koutsopoulos and M Titsias ldquoFirst learn thenearn optimizing mobile crowdsensing campaigns through data-drivenuser profilingrdquo in MobiHoc 2016

[227] D Yang G Xue X Fang and J Tang ldquoCrowdsourcing to smartphonesincentive mechanism design for mobile phone sensingrdquo in Proceedingsof the ACM International Conference on Mobile Computing andNetworking 2012 pp 173ndash184

[228] Ouml Yuumlruumlr C H Liu Z Sheng V C M Leung W Moreno and K KLeung ldquoContext-awareness for mobile sensing A survey and futuredirectionsrdquo IEEE Communications Surveys Tutorials vol 18 no 1 pp68ndash93 First Quarter 2016

[229] O Yurur and C H Liu Generic and energy-efficient context-awaremobile sensing CRC Press 2015

[230] S Taleb H Hajj and Z Dawy ldquoVCAMS Viterbi-based context awaremobile sensing to trade-off energy and delayrdquo IEEE Transactions onMobile Computing vol 17 no 1 pp 225ndash242 Jan 2018

[231] A Ahmad M M Rathore A Paul W-H Hong and H Seo ldquoContext-aware mobile sensors for sensing discrete events in smart environmentrdquoJournal of Sensors 2016

[232] X Hu X Li E C Ngai V C M Leung and P KruchtenldquoMultidimensional context-aware social network architecture for mobilecrowdsensingrdquo IEEE Communications Magazine vol 52 no 6 pp78ndash87 Jun 2014

[233] P Nguyen and K Nahrstedt ldquoContext-aware crowd-sensing in oppor-tunistic mobile social networksrdquo in IEEE 12th International Conferenceon Mobile Ad Hoc and Sensor Systems Oct 2015 pp 477ndash478

[234] J Wannenburg and R Malekian ldquoPhysical activity recognition fromsmartphone accelerometer data for user context awareness sensingrdquo




IEEE Transactions on Systems Man and Cybernetics Systems vol 47no 12 pp 3142ndash3149 Dec 2017

[235] S Faye R Frank and T Engel ldquoAdaptive activity and contextrecognition using multimodal sensors in smart devicesrdquo in InternationalConference on Mobile Computing Applications and Services Springer2015 pp 33ndash50

[236] H To L Fan L Tran and C Shahabi ldquoReal-time task assignment inhyperlocal spatial crowdsourcing under budget constraintsrdquo in IEEEInternational Conference on Pervasive Computing and Communications(PerCom) Mar 2016 pp 1ndash8

[237] E Wang Y Yang J Wu W Liu and X Wang ldquoAn efficient prediction-based user recruitment for mobile crowdsensingrdquo IEEE Transactionson Mobile Computing vol 17 no 1 pp 16ndash28 Jan 2018

[238] Q Xu and R Zheng ldquoWhen data acquisition meets data analyticsA distributed active learning framework for optimal budgeted mobilecrowdsensingrdquo in IEEE Conference on Computer Communications(INFOCOM) 2017 pp 1ndash9

[239] B Song H Shah-Mansouri and V W S Wong ldquoQuality of sensingaware budget feasible mechanism for mobile crowdsensingrdquo IEEETransactions on Wireless Communications vol 16 no 6 pp 3619ndash3631 Jun 2017

[240] Y Qu S Tang C Dong P Li S Guo H Dai and F Wu ldquoPostedpricing for chance constrained robust crowdsensingrdquo IEEE Transactionson Mobile Computing pp 1ndash1 2018

[241] Z He J Cao and X Liu ldquoHigh quality participant recruitmentin vehicle-based crowdsourcing using predictable mobilityrdquo in IEEEConference on Computer Communications (INFOCOM) Apr 2015 pp2542ndash2550

[242] G Cardone A Corradi L Foschini and R Ianniello ldquoParticipAct Alarge-scale crowdsensing platformrdquo IEEE Transactions on EmergingTopics in Computing vol 4 no 1 pp 21ndash32 Jan 2016

[243] R Rouvoy ldquoAPISENSE Mobile crowd-sensing made easyrdquo Jun 2016keynote of ECAASrsquo16 [Online] Available httpshalinriafrhal-01332594

[244] ldquoSenseMyCity Crowdsourcing an urban sensorrdquo 2014 [Online]Available httpcloudfuturecitiesupptsensemycity

[245] F Kalim J P Jeong and M U Ilyas ldquoCRATER A crowd sensingapplication to estimate road conditionsrdquo IEEE Access vol 4 pp 8317ndash8326 2016

[246] M-R Ra B Liu T F La Porta and R Govindan ldquoMedusa A pro-gramming framework for crowd-sensing applicationsrdquo in Proceedingsof the ACM International Conference on Mobile Systems Applicationsand Services 2012 pp 337ndash350

[247] T Das P Mohan V N Padmanabhan R Ramjee and A SharmaldquoPRISM platform for remote sensing using smartphonesrdquo in Proceedingsof the ACM International Conference on Mobile Systems Applicationsand Services 2010 pp 63ndash76

[248] I Carreras D Miorandi A Tamilin E R Ssebaggala and N ConcildquoMatador Mobile task detector for context-aware crowd-sensing cam-paignsrdquo in IEEE International Conference on Pervasive Computingand Communications Workshops (PERCOM Workshops) Mar 2013 pp212ndash217

[249] K Mehdi M Lounis A Bounceur and T Kechadi ldquoCupCarbonA multi-agent and discrete event wireless sensor network design andsimulation toolrdquo in International ICST Conference on Simulation Toolsand Techniques ser SIMUTools 2014 pp 126ndash131

[250] K Farkas and I Lendaacutek ldquoSimulation environment for investigatingcrowd-sensing based urban parkingrdquo in International Conference onModels and Technologies for Intelligent Transportation Systems (MT-ITS) Jun 2015 pp 320ndash327

[251] J Scott R Gass J Crowcroft P Hui C Diot and A ChaintreauldquoCRAWDAD dataset cambridgehaggle (v 2009-05-29)rdquo Downloadedfrom httpcrawdadorgcambridgehaggle20090529 May 2009

[252] N Eagle and A (Sandy) Pentland ldquoReality mining Sensing complexsocial systemsrdquo Personal Ubiquitous Comput vol 10 no 4 pp 255ndash268 Mar 2006

[253] N Kiukkonen B J O Dousse D Gatica-Perez and J K LaurilaldquoTowards rich mobile phone datasets Lausanne data collection campaignrdquoin ACM International Conference on Pervasive Services (ICPS) Jul2010

[254] L Aoude Z Dawy S Sharafeddine K Frenn and K Jahed ldquoSocial-aware device-to-device offloading based on experimental mobilityand content similarity modelsrdquo Wireless Communications and MobileComputing 2018

[255] Y Xiao P Simoens P Pillai K Ha and M Satyanarayanan ldquoLoweringthe barriers to large-scale mobile crowdsensingrdquo in Proc of the

ACM Workshop on Mobile Computing Systems and Applications serHotMobile 2013 pp 1ndash6

[256] K Farkas and I Lendaacutek ldquoEvaluation of simulation engines forcrowdsensing activitiesrdquo in International Conference amp WorkshopMechatronics in Practice and Education ser MECHEDU 2015 pp126ndash131

[257] G Cacciatore C Fiandrino D Kliazovich F Granelli and P BouvryldquoCost analysis of smart lighting solutions for smart citiesrdquo in IEEEInternational Conference on Communications (ICC) May 2017 pp1ndash6

[258] S Chessa M Girolami L Foschini R Ianniello A Corradi andP Bellavista ldquoMobile crowd sensing management with the ParticipActliving labrdquo Pervasive and Mobile Computing pp ndash 2016

[259] Z Zheng Y Peng F Wu S Tang and G Chen ldquoTrading data inthe crowd Profit-driven data acquisition for mobile crowdsensingrdquoIEEE Journal on Selected Areas in Communications vol 35 no 2 pp486ndash501 Feb 2017

[260] M Dai Z Su Y Wang and Q Xu ldquoContract theory based incentivescheme for mobile crowd sensing networksrdquo in Proc MoWNeT Jun2018 pp 1ndash5

[261] L Duan T Kubo K Sugiyama J Huang T Hasegawa and J WalrandldquoIncentive mechanisms for smartphone collaboration in data acquisitionand distributed computingrdquo in Proceedings IEEE INFOCOM Mar 2012pp 1701ndash1709

[262] C Jiang L Gao L Duan and J Huang ldquoScalable mobile crowdsensingvia Peer-to-Peer data sharingrdquo IEEE Transactions on Mobile Computingvol 17 no 4 pp 898ndash912 Apr 2018

[263] X Zeng L Gao C Jiang T Wang J Liu and B Zou ldquoA hybridpricing mechanism for data sharing in P2P-based mobile crowdsensingrdquoin 16th International Symposium on Modeling and Optimization inMobile Ad Hoc and Wireless Networks (WiOpt) May 2018 pp 1ndash8

[264] Y Li C A Courcoubetis and L Duan ldquoDynamic routing for social in-formation sharingrdquo IEEE Journal on Selected Areas in Communicationsvol 35 no 3 pp 571ndash585 Mar 2017

[265] M H Cheung F Hou and J Huang ldquoDelay-sensitive mobilecrowdsensing Algorithm design and economicsrdquo IEEE Transactionson Mobile Computing vol 17 no 12 pp 2761ndash2774 Dec 2018

[266] C Jiang L Gao L Duan and J Huang ldquoEconomics of Peer-to-Peermobile crowdsensingrdquo in IEEE Global Communications Conference(GLOBECOM) Dec 2015 pp 1ndash6

[267] Q Ma L Gao Y Liu and J Huang ldquoIncentivizing Wi-Fi networkcrowdsourcing A contract theoretic approachrdquo IEEEACM Transactionson Networking vol 26 no 3 pp 1035ndash1048 Jun 2018

[268] L Shu Y Chen Z Huo N Bergmann and L Wang ldquoWhen mobilecrowd sensing meets traditional industryrdquo IEEE Access vol 5 pp15 300ndash15 307 2017

[269] N Haderer R Rouvoy and L Seinturier ldquoA preliminary investigationof user incentives to leverage crowdsensing activitiesrdquo in IEEEInternational Conference on Pervasive Computing and CommunicationsWorkshops (PERCOM Workshops) 2013 pp 199ndash204

[270] T Luo H P Tan and L Xia ldquoProfit-maximizing incentive for partic-ipatory sensingrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2014 pp 127ndash135

[271] I Koutsopoulos ldquoOptimal incentive-driven design of participatorysensing systemsrdquo in IEEE Conference on Computer Communications(INFOCOM) Apr 2013 pp 1402ndash1410

[272] J Sun ldquoIncentive mechanisms for mobile crowd sensing Current statesand challenges of workrdquo arXiv preprint arXiv13108364 2013

[273] M Pouryazdan C Fiandrino B Kantarci T Soyata D Kliazovich andP Bouvry ldquoIntelligent gaming for mobile crowd-sensing participants toacquire trustworthy big data in the Internet of Thingsrdquo IEEE Accessvol 5 pp 22 209ndash22 223 Oct 2017

[274] T Tsujimori N Thepvilojanapong Y Ohta H Wang Y Zhao andY Tobe ldquoSenseUtil Utility vs incentives in participatory sensingrdquo inProc IEICE Workshop on Smart Sens Wireless Commun HumanProbes 2013 pp 24ndash29

[275] D Zhao X Y Li and H Ma ldquoBudget-feasible online incentive mech-anisms for crowdsourcing tasks truthfullyrdquo IEEEACM Transactions onNetworking vol 24 no 2 pp 647ndash661 Apr 2016

[276] S Reddy D Estrin and M Srivastava ldquoRecruitment frameworkfor participatory sensing data collectionsrdquo in Pervasive ComputingSpringer 2010 pp 138ndash155

[277] C Fiandrino F Anjomshoa B Kantarci D Kliazovich P Bouvryand J N Matthews ldquoSociability-driven framework for data acquisitionin mobile crowdsensing over fog computing platforms for smart citiesrdquoIEEE Transactions on Sustainable Computing vol 2 no 4 pp 345ndash358Oct 2017




[278] M Marjanovic A Antonic and I P Žarko ldquoEdge computingarchitecture for mobile crowdsensingrdquo IEEE Access vol 6 pp 10 662ndash10 674 2018

[279] P Vitello A Capponi C Fiandrino P Giaccone D KliazovichU Sorger and P Bouvry ldquoCollaborative data delivery for smart city-oriented mobile crowdsensing systemsrdquo in IEEE Global Communica-tions Conference (GLOBECOM) Dec 2018 pp 1ndash6

[280] J D Benedetto P Bellavista and L Foschini ldquoProximity discoveryand data dissemination for mobile crowd sensing using LTE directrdquoComputer Networks vol 129 pp 510 ndash 521 2017

[281] J Weppner and P Lukowicz ldquoBluetooth based collaborative crowd den-sity estimation with mobile phonesrdquo in IEEE International Conferenceon Pervasive Computing and Communications (PerCom) Mar 2013pp 193ndash200

[282] ldquoBluetooth specification version 4rdquo Bluetooth SIG Jul 2017 [Online]Available httpswwwbluetoothorgdocmanhandlersdownloaddocashxdoc_id=229737

[283] ldquoWaze Get the best route every day with realndashtime help from otherdriversrdquo httpswwwwazecom 2006

[284] H Habibzadeh Z Qin T Soyata and B Kantarci ldquoLarge-scaledistributed dedicated- and non-dedicated smart city sensing systemsrdquoIEEE Sensors Journal vol 17 no 23 pp 7649ndash7658 Dec 2017

[285] X Chen Y Chen M Dong and C Zhang ldquoDemystifying energy usagein smartphonesrdquo in ACMEDACIEEE Design Automation Conference(DAC) Jun 2014 pp 1ndash5

[286] N Ravi N Dandekar P Mysore and M L Littman ldquoActivityrecognition from accelerometer datardquo in AAAI vol 5 2005 pp 1541ndash1546

[287] A Bujari B Licar and C E Palazzi ldquoMovement pattern recognitionthrough smartphonersquos accelerometerrdquo in IEEE Consumer communica-tions and networking conference (CCNC) 2012 pp 502ndash506

[288] I Shafer and M L Chang ldquoMovement detection for power-efficientsmartphone WLAN localizationrdquo in Proceedings of the ACM interna-tional conference on Modeling analysis and simulation of wirelessand mobile systems 2010 pp 81ndash90

[289] P Siirtola and J Roumlning ldquoRecognizing human activities user-independently on smartphones based on accelerometer datardquo Inter-national Journal of Interactive Multimedia and Artificial Intelligencevol 1 no 5 2012

[290] A Anjum and M U Ilyas ldquoActivity recognition using smartphone sen-sorsrdquo in IEEE Consumer Communications and Networking Conference(CCNC) 2013 pp 914ndash919

[291] M Shoaib H Scholten and P J Havinga ldquoTowards physical activityrecognition using smartphone sensorsrdquo in IEEE International Confer-ence on Ubiquitous Intelligence and Computing and on Autonomic andTrusted Computing (UICATC) 2013 pp 80ndash87

[292] C Schuss T Leikanger B Eichberger and T Rahkonen ldquoEfficient useof solar chargers with the help of ambient light sensors on smartphonesrdquoin IEEE Conference of Open Innovations Association (FRUCT) 2014pp 79ndash85

[293] V Freschi S Delpriori E Lattanzi and A Bogliolo ldquoBootstrap baseduncertainty propagation for data quality estimation in crowdsensingsystemsrdquo IEEE Access vol 5 pp 1146ndash1155 2017

[294] M Tomasoni A Capponi C Fiandrino D Kliazovich F Granelliand P Bouvry ldquoWhy energy matters profiling energy consumptionof mobile crowdsensing data collection frameworksrdquo Pervasive andMobile Computing vol 51 pp 193 ndash 208 2018 [Online] AvailablehttpwwwsciencedirectcomsciencearticlepiiS1574119217305965

[295] X Sheng J Tang and W Zhang ldquoEnergy-efficient collaborative sensingwith mobile phonesrdquo in IEEE Conference on Computer Communications(INFOCOM) Mar 2012 pp 1916ndash1924

[296] H Xiong D Zhang L Wang and H Chaouchi ldquoEMC3 Energy-efficient data transfer in mobile crowdsensing under full coverageconstraintrdquo IEEE Transactions on Mobile Computing vol 14 no 7pp 1355ndash1368 2015

[297] C Zhang K Subbu J Luo and J Wu ldquoGROPING Geomagnetismand crowdsensing powered indoor navigationrdquo IEEE Transactions onMobile Computing vol 14 no 2 pp 387ndash400 Feb 2015

[298] L Zhang J Liu H Jiang and Y Guan ldquoSensTrack Energy-efficientlocation tracking with smartphone sensorsrdquo IEEE Sensors Journalvol 13 no 10 pp 3775ndash3784 Oct 2013

[299] E Ozer and M Q Feng ldquoDirection-sensitive smart monitoring ofstructures using heterogeneous smartphone sensor data and coordinatesystem transformationrdquo Smart Materials and Structures vol 26 no 4p 045026 2017

[300] V M Souza R Giusti and A J Batista ldquoAsfault A low-cost systemto evaluate pavement conditions in real-time using smartphones and

machine learningrdquo Pervasive and Mobile Computing vol 51 pp 121 ndash137 2018 [Online] Available httpwwwsciencedirectcomsciencearticlepiiS1574119218300518

[301] T Ludwig C Reuter T Siebigteroth and V Pipek ldquoCrowdMonitorMobile crowd sensing for assessing physical and digital activities ofcitizens during emergenciesrdquo in Proc of the ACM Conference on HumanFactors in Computing Systems ser SIGCHI 2015 pp 4083ndash4092

[302] N B Grimm S H Faeth N E Golubiewski C L Redman J WuX Bai and J M Briggs ldquoGlobal change and the ecology of citiesrdquo inScience vol 319 no 5864 2008 pp 756ndash760

[303] H B Dulal and S Akbar ldquoGreenhouse gas emission reduction optionsfor cities Finding the ldquocoincidence of agendasrdquo between local prioritiesand climate change mitigation objectivesrdquo Habitat International vol 38pp 100 ndash 105 2013

[304] A Caragliu C Del Bo and P Nijkamp ldquoSmart cities in europerdquoJournal of urban technology vol 18 no 2 pp 65ndash82 2011

[305] A Longo M Zappatore M Bochicchio and S B Navathe ldquoCrowd-sourced data collection for urban monitoring via mobile sensorsrdquo ACMTrans Internet Technol vol 18 no 1 pp 1ndash21 Oct 2017

[306] Y Zhou B P L Lau C Yuen B Tunccediler and E Wilhelm ldquoUnder-stand urban human mobility through crowdsensed datardquo CoRR volabs180500628 2018

[307] M S Kaiser K T Lwin M Mahmud D Hajializadeh T Chaipimon-plin A Sarhan and M A Hossain ldquoAdvances in crowd analysis forurban applications through urban event detectionrdquo IEEE Transactionson Intelligent Transportation Systems pp 1ndash21 2017

[308] ldquoCisco global cloud index Forecast and methodology 2016ndash2021rdquoCisco Visual Networking 2018 White Paper

[309] X Cheng L Fang X Hong and L Yang ldquoExploiting mobile big dataSources features and applicationsrdquo IEEE Network vol 31 no 1 pp72ndash79 Jan 2017

[310] Y Sun H Song A J Jara and R Bie ldquoInternet of Things and bigdata analytics for smart and connected communitiesrdquo IEEE Accessvol 4 pp 766ndash773 2016

[311] C Zhang P Patras and H Haddadi ldquoDeep learning in mobile andwireless networking A surveyrdquo CoRR vol abs180304311 2018[Online] Available httparxivorgabs180304311

[312] J Wang Y Wang D Zhang J Goncalves D Ferreira A Visuri andS Ma ldquoLearning-assisted optimization in mobile crowd sensing Asurveyrdquo IEEE Transactions on Industrial Informatics vol 15 no 1pp 15ndash22 Jan 2019

[313] Y LeCun Y Bengio and G Hinton ldquoDeep learningrdquo Nature vol 521no 7553 p 436 2015

[314] A Dubey N Naik D Parikh R Raskar and C A Hidalgo ldquoDeeplearning the city Quantifying urban perception at a global scalerdquo inComputer Vision ndash ECCV Springer International Publishing 2016pp 196ndash212

[315] ldquoFoobot Better air better liferdquo 2017 [Online] Available httpsfoobotio

[316] Y Lv Y Duan W Kang Z Li and F Y Wang ldquoTraffic flow predictionwith big data A deep learning approachrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 2 pp 865ndash873 Apr2015

[317] ldquoDangerous by designrdquo 2011 [Online] Available httpt4americaorgdocsdbd2011Dangerous-by-Design-2011pdf

[318] L Wang D Zhang D Yang A Pathak C Chen X Han H Xiongand Y Wang ldquoSPACE-TA Cost-effective task allocation exploitingintradata and interdata correlations in sparse crowdsensingrdquo ACM TransIntell Syst Technol vol 9 no 2 pp 1ndash20 Oct 2017

[319] L Fu S Ma L Kong S Liang and X Wang ldquoFINE A frameworkfor distributed learning on incomplete observations for heterogeneouscrowdsensing networksrdquo IEEEACM Transactions on Networkingvol 26 no 3 pp 1092ndash1109 Jun 2018

[320] S Yang K Han Z Zheng S Tang and F Wu ldquoTowards personalizedtask matching in mobile crowdsensing via fine-grained user profilingrdquoin IEEE Conference on Computer Communications (INFOCOM) Apr2018

[321] K Christidis and M Devetsikiotis ldquoBlockchains and smart contractsfor the Internet of Thingsrdquo IEEE Access vol 4 pp 2292ndash2303 2016

[322] ldquoHyperledgerrdquo 2016 [Online] Available httpswwwhyperledgerorg[323] G Wood ldquoEthereum A secure decentralised generalised transaction

ledgerrdquo Ethereum project yellow paper vol 151 pp 1ndash32 2014[324] P Hu S Dhelim H Ning and T Qiu ldquoSurvey on fog computing

architecture key technologies applications and open issuesrdquo Journalof Network and Computer Applications vol 98 pp 27 ndash 42 2017




[325] C Fiandrino N Allio D Kliazovich P Giaccone and P BouvryldquoProfiling performance of application partitioning for wearable devicesin mobile cloud and fog computingrdquo IEEE Access vol 7 pp 12 156ndash12 166 Jan 2019

[326] P P Jayaraman J B Gomes H L Nguyen Z S Abdallah S Krish-naswamy and A Zaslavsky ldquoScalable energy-efficient distributed dataanalytics for crowdsensing applications in mobile environmentsrdquo IEEETrans on Computational Social Systems vol 23 pp 109ndash123 Sep2015

[327] P Bellavista S Chessa L Foschini L Gioia and M Girolami ldquoHuman-enabled edge computing Exploiting the crowd as a dynamic extensionof mobile edge computingrdquo IEEE Communications Magazine vol 56no 1 pp 145ndash155 Jan 2018

[328] R Roman J Lopez and M Mambo ldquoMobile edge computing fog etal A survey and analysis of security threats and challengesrdquo FutureGeneration Computer Systems vol 78 pp 680 ndash 698 2018

[329] F Bonomi R Milito J Zhu and S Addepalli ldquoFog computing and itsrole in the Internet of Thingsrdquo in Proceedings of the ACM Workshopon Mobile Cloud Computing ser MCC 2012 pp 13ndash16

[330] T Taleb K Samdanis B Mada H Flinck S Dutta and D SabellaldquoOn multi-access edge computing A survey of the emerging 5G networkedge cloud architecture and orchestrationrdquo IEEE CommunicationsSurveys Tutorials vol 19 no 3 pp 1657ndash1681 Third Quarter 2017

[331] Z Zhou H Liao B Gu K M S Huq S Mumtaz and J RodriguezldquoRobust mobile crowd sensing When deep learning meets edgecomputingrdquo IEEE Network vol 32 no 4 pp 54ndash60 Jul 2018

[332] S Yang J Bian L Wang H Zhu Y Fu and H Xiong ldquoEdgesenseEdge-mediated spatial-temporal crowdsensingrdquo IEEE Access pp 1ndash12018

[333] M Shafi A F Molisch P J Smith T Haustein P Zhu P D SilvaF Tufvesson A Benjebbour and G Wunder ldquo5G A tutorial overviewof standards trials challenges deployment and practicerdquo IEEE Journalon Selected Areas in Communications vol 35 no 6 pp 1201ndash1221Jun 2017

[334] M Simsek A Aijaz M Dohler J Sachs and G Fettweis ldquo5G-EnabledTactile Internetrdquo IEEE Journal on Selected Areas in Communicationsvol 34 no 3 pp 460ndash473 Mar 2016

[335] C Fiandrino H Assasa P Casari and J Widmer ldquoScaling millimeter-wave networks to dense deployments and dynamic environmentsrdquoProceedings of the IEEE vol 107 no 4 pp 732ndash745 Apr 2019

Andrea Capponi (Srsquo16) is a PhD candidate atthe University of Luxembourg He received theBachelor Degree and the Master Degree both inTelecommunication Engineering from the Universityof Pisa Italy His research interests are mobilecrowdsensing urban computing and IoT

Claudio Fiandrino (Srsquo14-Mrsquo17) is a postdoctoral re-searcher at IMDEA Networks He received his PhDdegree at the University of Luxembourg (2016) theBachelor Degree (2010) and Master Degree (2012)both from Politecnico di Torino (Italy) Claudio wasawarded with the Spanish Juan de la Cierva grantand the Best Paper Awards in IEEE CloudNetrsquo16and in ACM WiNTECHrsquo18 His research interestsinclude mobile crowdsensing edge computing andURLLC

Burak Kantarci (Srsquo05ndashMrsquo09ndashSMrsquo12) is an Asso-ciate Professor and the founding director of theNext Generation Communications and ComputingNetworks (NEXTCON) Research laboratory in theSchool of Electrical Engineering and ComputerScience at the University of Ottawa (Canada) Hereceived the MSc and PhD degrees in computerengineering from Istanbul Technical University in2005 and 2009 respectively He is an Area Editorof the IEEE TRANSACTIONS ON GREEN COMMU-NICATIONS NETWORKING an Associate Editor of

the IEEE COMMUNICATIONS SURVEYS AND TUTORIALS and an AssociateEditor of the IEEE ACCESS and IEEE Networking Letters He also servesas the Chair of the IEEE ComSoc Communication Systems Integration andModeling Technical Committee Dr Kantarci is a senior member of the IEEEmember of the ACM and a distinguished speaker of the ACM

Luca Foschini (Mrsquo04-SMrsquo19) graduated from theUniversity of Bologna from which he received aPhD degree in Computer Engineering in 2007 Heis now an Assistant Professor of Computer Engi-neering at the University of Bologna His interestsinclude distributed systems and solutions for systemand service management management of cloudcomputing context data distribution platforms forsmart city scenarios context-aware session controlmobile edge computing and mobile crowdsensingand crowdsourcing He is a member of ACM and

IEEE

Dzmitry Kliazovich (Mrsquo03-SMrsquo12) is a Head ofInnovation at ExaMotive Dr Kliazovich holds anaward-winning PhD in Information and Telecommu-nication Technologies from the University of Trento(Italy) He coordinated organization and chaired anumber of highly ranked international conferencesand symposia Dr Kliazovich is a frequent keynoteand panel speaker He is Associate Editor of IEEECommunications Surveys and Tutorials and of IEEETransactions of Cloud Computing His main researchactivities are in intelligent and shared mobility energy

efficiency cloud systems and architectures data analytics and IoT

Pascal Bouvry is a professor at the Faculty ofScience Technology and Communication at theUniversity of Luxembourg and faculty member atSnT Center Prof Bouvry has a PhD in computerscience from the University of Grenoble (INPG)France He is on the IEEE Cloud Computing andElsevier Swarm and Evolutionary Computation edi-torial boards communication vice-chair of the IEEESTC on Sustainable Computing and co-founder ofthe IEEE TC on Cybernetics for Cyber-PhysicalSystems His research interests include cloud amp

parallel computing optimization security and reliability




SensingLayer

CommunicationLayer

DataLayer

ApplicationLayer

Fig 1 Layered architecture of MCS systems It shows the four-layered architecture that describes the MCS paradigm including application (Fig 5) data(Fig 6) communication (Fig 7) and sensing (Fig 8) layers which are discussed in Sec III-A

aim at using ICT solutions to improve the management ofeveryday life of their citizens [27] [28] The Internet ofThings (IoT) paradigm is the candidate solution for a widedeployment of sensing infrastructure enabling smart citiesrsquoapplications [29] Moreover active participation of citizenscan improve spatial coverage of already deployed sensingsystems with no need for further investments MCS leverageshuman intelligence which has a deeper context understandingthan traditional sensor networks Cities are facing significantdeficits in infrastructure services and humans can be involvedto improve their monitoring and maintenance We illustratethis concept with specific use cases Data harvested fromsmartphonesrsquo accelerometers over moving vehicles enablesthe detection of bridge vibrations [30] Other possible cityservices where MCS plays a fundamental role are smarttraffic management [31] [32] or free parking spot detectionSpecifically ParkSense detects vacant parking spots using WiFiscans of smartphones [33] while ParkGauge reports real-timecrowdsensed information about indoor parking occupancy andexploits low-consuming sensors (eg accelerometer) to detectthe driving states [34]

There is a wealth of literature on MCS systems Considerableresearch efforts proposed novel sensing architectures andas already mentioned investigated specific aspects such asincentive mechanisms privacy issues as well as the reliabilityand trust of the data collection process With such a large bodyof work surveys on the topic covering specific areas such asprivacy [15] or incentives for user recruitment [12] [13] already

exists Others like [35] present many aspects components ofMCS as an emerging paradigm eg data collection qualityassurance etc However this is an early work missing a wealthof early developments in the field What is missing in thispicture is a survey that covers the whole decade of research inMCS and provides a clear state of its evolution This is oneof the purposes of our manuscript Also the vast amount ofwork in literature remains uncategorized with many of thecore paradigms unclear To illustrate for example there isno consensus on the term ldquoopportunistic sensingrdquo Accordingto Ganti et al [7] ldquoopportunistic sensingrdquo is defined as ldquoOnthe other hand opportunistic sensing is where the sensingis more autonomous and user involvement is minimal (egcontinuous location sampling)rdquo However Khan et al [9] statethat opportunistic sensing requires no user involvement at allsince the decisions to perform the sensing are a prerogative ofthe device itself Finally Han et al [36] enlarge the previousvision of opportunistic sensing in the context of single-userinvolvement and they describe opportunistic sensing as aparadigm enabling cooperation among smartphones Typicallyboth terms ldquoopportunisticrdquo and ldquoparticipatoryrdquo sensing remainunder the common umbrella of MCS [7] [9] [36] [37] In othercases both ldquomobile crowdsensingrdquo and ldquoparticipatory sensingrdquoare used interchangeably [31] Other times both ldquomobilecrowdsensingrdquo ldquoparticipatory sensingrdquo and ldquoopportunisticsensingrdquo are synonyms [38]

With this plethora of definitions our manuscript has theprecise objective of proposing detailed taxonomies to sim-




Sec IIBackground

Related surveysIIA

TimelineIIB


IIC


IID

Featureextraction



IIIB


IIIA


IIIC


IIID


Final RemarksIIIF



TaxonomiesIVA

ClassificationIVB

Task

User


on Data Layer

TaxonomiesVA

ClassificationVB

Management

Processing



TaxonomiesVIA

ClassificationVIB

Technologies

Reporting


on Sensing Layer

TaxonomiesVIIA

ClassificationVIIB

Elements

Samplingprocess

Sec VIIIDiscussion



















II BACKGROUND


A Related Surveys














[35] [40] [41] [42]ndash[44]


[45]ndash[49]


[8] [9]


[50] [51]


[12] [13] [52] [53]


[15] [16]




B Timeline















2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]







MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS


campaigns [67]





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE








Sec IIBackground

Related surveysIIA

TimelineIIB


IIC


IID

Featureextraction



IIIB


IIIA


IIIC


IIID


Final RemarksIIIF



TaxonomiesIVA

ClassificationIVB

Task

User


on Data Layer

TaxonomiesVA

ClassificationVB

Management

Processing



TaxonomiesVIA

ClassificationVIB

Technologies

Reporting


on Sensing Layer

TaxonomiesVIIA

ClassificationVIIB

Elements

Samplingprocess

Sec VIIIDiscussion



















II BACKGROUND


A Related Surveys














[35] [40] [41] [42]ndash[44]


[45]ndash[49]


[8] [9]


[50] [51]


[12] [13] [52] [53]


[15] [16]




B Timeline















2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]







MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS


campaigns [67]





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE









II BACKGROUND


A Related Surveys














[35] [40] [41] [42]ndash[44]


[45]ndash[49]


[8] [9]


[50] [51]


[12] [13] [52] [53]


[15] [16]




B Timeline















2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]







MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS


campaigns [67]





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE











[35] [40] [41] [42]ndash[44]


[45]ndash[49]


[8] [9]


[50] [51]


[12] [13] [52] [53]


[15] [16]




B Timeline















2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]







MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS


campaigns [67]





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE








2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

[56][57] [18]

[58] [59]

[7]

[6]

[15]

[60]

[62][8]

[36]

[61] [63]

[64]

[9]

[14]

[65]

[34]

[66] [67]







MOBILE

CROWDSENSING

MOBILEPHONE

SENSING

CROWDSOURCING

SMART DEVICESamp

WEARABLES

HUMANFACTOR

CLOUDCOMPUTING

INTERNETOF

THINGS

WIRELESSSENSOR

NETWORKS


campaigns [67]





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE





















[109]


[99]


[95]

















BP





Fog Fog

Raw Data

Cloud






Reporting to GO

Bluetooth

WiFi-Direct

WiFi

Cellular

Group Owner (GO)

Group User

Single User

Base station




WiFi



Smart Devices





















[24] [117]ndash[120]


[21] [22] [56] [115] [121]ndash[129]


[130]ndash[133]


[17] [19] [116] [134]ndash[136]


[137]ndash[139]


[20] [140]ndash[145]


[62] [146]ndash[155]


[156] [157]


[158]ndash[160]


[30] [161] [162]


[25] [26]


[163]ndash[165]


[147] [148] [166]ndash[169]



























[172] [173]


[174]ndash[178]


[179]ndash[181]


[182] [183]


[184]ndash[187]


[188]ndash[193]




























[36] [195]ndash[197]


[198]ndash[205]


[206]ndash[217]


[218]ndash[227]


[228]ndash[235]


[112] [236]ndash[239] [240]































[242]


[243]


[244]


[245]


[246]


[247]


[63]


[248]


[68]


[66]


[249]


[250]


[64]


[251]


[252]


[253]


[254]



















ApplicationLayer

Task

User


Processing

CommunicationLayer

Technologies

Reporting


Elements






A Taxonomies















Application Layer

Task

User

Assignment

Scheduling

Execution

Type

Recruitment

Selection

Pro-active

Reactive

Central authority

Decentralized

Single task

Multi-tasking

Voluntary

Incentivized

Consumer

Contributor

Platform centric

User centric





Cloud Collector

alloc

User Recruitment

allo

alloc

Task Allocation

allo

alloc













B Classification



























A Taxonomies












Data Layer

Management

Processing

Format

Storage

Dimension

Analytics

Pre-processing

Post-processing

Centralized

Distributed

Structured

Unstructured

Single dimension

Multi-dimensional

Raw data


ML amp data mining

Real-time

Statistical

Prediction

















B Classification
































A Taxonomies















CommunicationLayer

Technologies

Reporting

Infrastructured

Infrastructure-less

Methodology

Upload mode

Timing

Cellular

WLAN

WiFi-Direct

LTE-Direct

Bluetooth

Relay

Store amp forward

Individual

Collaborative

Synchrhonous

Asynchrhonous









B Classification
































A Taxonomies













Sensing Layer

Elements

Sampling Process

Activity

Deployment

Acquisition

Responsibility

Frequency

Userinvolvement

Dedicated

Non-dedicated

Always-on

On-demand

Homogeneous

Heterogeneous

Continuous

Event-based

Central collector

Mobile devices

Participatory

Opportunistic
















B Classification































VIII DISCUSSION





MOBILE

CROWDSENSING


EDGE COMPUTING

DISTRIBUTEDPAYMENT

PLATFORMS

BIG DATAMACHINEDEEP


ANDURBAN

COMPUTING

5G NETWORKS
























REFERENCES





















































































































































































































































































































































































IEEE







ieee communications surveys & tutorials 1 a survey on

Documents