Developing Vehicular Data CloudServices in the IoT Environment

Wu He, Gongjun Yan, and Li Da Xu, Senior Member, IEEE

Abstract—The advances in cloud computing and internet ofthings (IoT) have provided a promising opportunity to resolve thechallenges caused by the increasing transportation issues. Wepresent a novelmultilayered vehicular data cloud platformby usingcloud computing and IoT technologies. Two innovative vehiculardata cloud services, an intelligent parking cloud service and avehicular data mining cloud service, for vehicle warranty analysisin the IoT environment are also presented. Two modified datamining models for the vehicular data mining cloud service, a NaïveBayes model and a Logistic Regression model, are presented indetail. Challenges and directions for future work are also provided.

Index Terms—Automobile service, cloud computing, internet ofthings (IoT), intelligent transportation systems (ITSs), service-oriented architecture (SOA).

I. INTRODUCTION

M ODERN VEHICLES are increasingly equipped with alarge amount of sensors, actuators, and communication

devices (mobile devices,GPS devices, and embedded computers).In particular, numerous vehicles have possessed powerful

sensing, networking, communication, and data processing capa-bilities, and can communicate with other vehicles or exchangeinformation with the external environments over various proto-cols, including HTTP, TCP/IP, SMTP, WAP, and Next Genera-tionTelematics Protocol (NGTP) [1].As a result,many innovativetelematics services [2], such as remote security for disablingengine and remote diagnosis, have been developed to enhancedrivers’ safety, convenience, and enjoyment.

The advances in cloud computing and internet of things (IoT)have provided a promising opportunity to further address theincreasing transportation issues, such as heavy traffic, conges-tion, and vehicle safety. In the past few years, researchers haveproposed a few models that use cloud computing for implement-ing intelligent transportation systems (ITSs). For example, a newvehicular cloud architecture called ITS-Cloud was proposed toimprove vehicle-to-vehicle communication and road safety [3].

A cloud-based urban traffic control system was proposed tooptimize traffic control [4]. Based on a service-oriented archi-tecture (SOA), this system uses a number of software services(SaaS), such as intersection control services, area managementservice, cloud service discovery service, and sensor service, toperform different tasks. These services also interact with eachother to exchange information and provide a solid basis forbuilding a collaborative traffic control and processing system in adistributed cloud environment.

As an emerging technology caused by rapid advances inmodern wireless telecommunication, IoT has received a lot ofattention and is expected to bring benefits to numerous applica-tion areas including health care, manufacturing, and transporta-tion [5]–[8]. Currently, the use of IoT in transportation is still inits early stage and most research on ITSs has not leveraged theIoT technology as a solution or an enabling infrastructure. To thisend, we propose to use both cloud computing and IoT as anenabling infrastructure for developing a vehicular data cloudplatformwhere transportation-related information, such as trafficcontrol and management, car location tracking and monitoring,road condition, car warranty, and maintenance information, canbe intelligently connected and made available to drivers, auto-makers, part-manufacturer, vehicle quality controller, safetyauthorities, and regional transportation division. An experimentof using data mining models to analyze vehicular data clouds inthe IoT environment was also conducted to demonstrate thefeasibility of vehicular data mining service.

The rest of the paper is organized as follows. In Section II, weprovide a brief review of vehicular networks, cloud computing inthe automotive domain, and IoT in the automotive domain. InSection III, we propose a novel multilayered vehicular data cloudplatform using existing cloud computing and IoT technologies.Section IV presents two innovative vehicular data cloud services;an intelligent parking cloud service and a vehicular data miningcloud service for vehicle warranty analysis in the IoT environ-ment. Two modified data mining models for the vehicular datamining cloud service, a Naïve Bayes model and a LogisticRegression model, are presented in detail. Challenges and direc-tions for future work are given in Section V. Section VI presentsour conclusion.

II. RELATED WORK

A. Vehicular Networks

Wireless technology leads to the development of vehicularnetworks in the past decades. The original idea is that theroadside infrastructure and the radio-equipped vehicles couldcommunicate using wireless networks. To make networkingoperations such as routing more effective, researchers had

Manuscript received September 12, 2013; revised November 22, 2013;accepted January 02, 2014. Date of publication January 10, 2014; date ofcurrent version May 02, 2014. This work was supported in part by theNational Natural Science Foundation of China (NNSFC) under Grant71132008, and in part by U.S. National Science Foundation under GrantsSES-1318470 and 1044845. Paper no. TII-13-0629.

W. He is with Old Dominion University, Norfolk, VA 23529 USA (e-mail:whe@odu.edu).

G. Yan is with the University of Southern Indiana, Evansville, IN 47712USA(e-mail: gyan@usi.edu).

L. D. Xu is with the Institute of Computing Technology, Chinese Academy ofSciences, Beijing 100190, China; with Shanghai Jiao Tong University, Shanghai20024, China; with University of Science and Technology of China, Hefei230026, China; and also with Old Dominion University, Norfolk, VA 23529USA (e-mail: lxu@odu.edu).

Digital Object Identifier 10.1109/TII.2014.2299233

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014 1587

1551-3203 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

developed a dynamic inter-vehicle network called vehicularad-hoc networks (VANET). VANETs were primarily designedto support the communication between different vehicles (V2 V)and the communication between vehicles and the roadsideinfrastructures (V2I) [9]. VANETs possess hybrid architectureand integrate ad hoc networks, wireless LAN, and cellulartechnology [10] for ITS. Furthermore, many VANET applica-tions were developed by numerous vehicle manufacturers, gov-ernment agencies, and industrial organizations. Initially, mostVANET applications were focused on improving drivers’ safetyand offered functions such as traffic monitoring and update,emergency warning, and road assistance [11]. In recent years,many nonsafety-related VANET applications, such as entertain-ment and gaming applications, have been developed.

B. Cloud Computing in the Automotive Domain

Cloud computing has been proposed to reshape vehicularsoftware and services in the automotive domain. As more andmore cars are equipped with devices that can access the internet,Olariu et al. [11] propose to integrate existing vehicular net-works, various sensors, on-board devices in vehicles, and cloudcomputing to create vehicular clouds. They suggest that vehicu-lar clouds are technologically feasible and will have a significantimpact on the society once they are built. Thus, both existingautomobile software and a variety of information resources arebeing virtualized and packaged as services to build vehicularclouds. Different vehicular services are often combined and usedto implement the mapping, encapsulation, aggregation, andcomposition and allow vehicles to interact with various hostedservices outside the vehicles. Currently, using the modularapproach, multilayer and SOAs to integrate various vehicularresources and services appears to be the most promising modeland framework for building vehicular cloud service platforms.By using the modular approach to decompose a complex systeminto smaller subsystems according to their functions, we candivide a vehicular cloud service platform into a number offunctional services and subsystems such as traffic administration,service routing, information processing, vehicle warranty anal-ysis and mining, etc. As cloud computing includes three distinctservices—platform as a service (PaaS), infrastructure as a service(IaaS) as well as the popular software as a service (SaaS), acompound of SaaS, PaaS, and IaaS should be leveraged forbuilding vehicular cloud service platforms. Furthermore, cloudscan also be divided into private, public, and hybrid clouds. Thus,vehicular cloud service platforms can also be designed to be ahybrid cloud where some services, such as user informationquery, can be hosted on public cloud platforms and othermissing-critical services, such as traffic administration, shouldbe hosted on private cloud platforms [12]. A taxonomy wasdeveloped to classify VANET-related clouds into the followingthree types: 1) vehicles using clouds; 2) vehicular clouds; and3) hybrid clouds [13].

Multilayer approaches and SOA [14]–[16] have been proposedas the main architecture to construct various vehicular cloudservice platforms. Iwai and Aoyama [1] propose to develop acloud service system for automobiles (a.k.a., the DARWINsystem) using SOA as an enabling architecture [17], [18].

DARWIN contains key service components, such as ServiceProcess Manager and Service Space, and these componentsinteract with various services both inside and outside of vehiclesto form a comprehensive vehicular cloud. DARWIN also pro-vides protocols to support interoperability between existingvehicular software and cloud-based services. Wang et al. [19]propose a vehicle cloud computing architecture composed ofthree functional tiers: 1) cloud service; 2) communication; and3) device tiers. By using cloud computing techniques such asSOA, the three-layer architecture allows heterogeneous devices,network, and services to exchange information and collaborate ina real-time manner. A three-layer V-Cloud architecture wasproposed [10] to combine vehicular cyber-physical systems withcloud computing technologies to offer essential services fordrivers. The V-Cloud architecture includes three layers: in-carvehicular cyber-physical system, V2 V network, and V2I net-work. Each layer has numerous sub-components. The ITS-CloudproposedbyBitamandMellouk [3] includes three layers: 1) cloudlayer; 2) communication layer; and 3) end-users layer. In partic-ular, the cloud layer was divided into both static and dynamiccloud to support different services needed byvarious stakeholdersof the vehicular clouds. A new architecture named VehiCloudwas developed to transform traditional vehicular networks into aservice-oriented cloud architecture [20]. By taking advantage ofemerging cloud computing technologies [21]–[23], VehiCloudhas been implemented and tested to addressV2Vcommunicationissues and extend the capabilities of embedded devices andmobile devices though road experiments.

C. IoT in the Automotive Domain

The integration of sensors and communication technologiesprovides a way for us to track the changing status of an objectthrough the Internet. IoT explains a future in which a variety ofphysical objects and devices around us, such as various sensors,radio frequency identification (RFID) tags, GPS devices, andmobile devices, will be associated to the Internet and allows theseobjects and devices to connect, cooperate, and communicatewithin social, environmental, and user contexts to reach commongoals [24], [25]. As an emerging technology, the IoT is expectedto offer promising solutions to transform transportation systemsand automobile services in the automobile industry. Speed andShingleton [26] propose an idea to use the “unique identifyingproperties of car registration plates” to connect various things.Asvehicles have increasingly powerful sensing, networking, com-munication, and data processing capabilities, IoT technologiescan be used to harness these capabilities and share under-utilizedresources among vehicles in the parking space or on the road.For example, IoT technologies make it possible to track eachvehicle’s existing location, monitor its movement, and predict itsfuture location.

By integrating with cloud computing, wireless sensor net-work, RFID sensor networks, satellite network, and otherintelligent transportation technologies, a new generation ofIoT-based vehicular data clouds can be developed and deployedto bring many business benefits, such as predicting increasingroad safety, reducing road congestion, managing traffic, andrecommending car maintenance or repair. Some preliminary

1588 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

work of using IoT technologies to improve ITSs has beenconducted in recent years. For example, an intelligent informat-ics system (iDrive system) developed by BMW used varioussensors and tags tomonitor the environment, such as tracking thevehicle location and the road condition, to provide drivingdirections [27]. Leng and Zhao [12] propose an intelligentinternet-of-vehicles system (known as IIOVMS) to collect trafficinformation from the external environments on an ongoing basisand to monitor and manage road traffic in real time. Lumpkins[28] discusses how ITSs could use IoT devices in the vehicle toconnect to the cloud and hownumerous sensors on the road couldbe virtualized to leverage the processing capabilities of the cloud.Qin et al. [27] propose a technology architecture that uses cloudcomputing, IoT, and middleware technologies to enable theinnovation of automobile services. Zhang et al. [29] designedan intelligent monitoring system to track the location of refrig-erator trucks using IoT technologies.

III. PROPOSED VEHICULAR DATA CLOUD PLATFORMIN THE IOT ENVIRONMENT

Fig. 1 shows the layered architecture of our proposed IoT-based vehicular data cloud platform. By integrating variousdevices such as sensors, actuators, controllers, GPS devices,mobile phones, and other Internet access equipments, and em-ploying networking technologies (wireless sensor network, cel-lular network, satellite network, and others), cloud computing,IOT, and middleware, this platform supports V2 V and V2Icommunication mechanisms and is able to collect and exchangedata among the drivers, vehicles, and roadside infrastructuresuch as cameras and street lights. The goal of this platform is toprovide real-time, economic, secure, and on-demand services tocustomers through the associated clouds including a conven-tional cloud and a temporary cloud (vehicular cloud) [3]. Theconventional cloud is composed of virtualized computers andprovides SaaS, PaaS, and IaaS to interested customers. Forexample, cloud management services and many traffic adminis-tration applications can be hosted on the conventional cloud. Thetemporary cloud is typically formed on demand and is composedof under-utilized computing, networking, and storage facilitiesof vehicles and is designed to expand the conventional cloud inorder to increase the whole cloud’s computing, processing, andstoring capabilities. The temporary cloud supports a compoundof SaaS, PaaS, and IaaS and primarily hosts highly dynamicvehicular applications which may have issues running on theconventional clouds [26]. For example, traffic-related applica-tions and smart parking applications are suitable for the tempo-rary cloud. The temporary cloud often needs to communicatewith the conventional clouds and there is a frequent exchange ofdata and services between the two clouds [13]. Based on thelayered architecture in Fig. 1, heterogeneous IoT-related devices,network, community technologies, and cloud-based services ondifferent layers can be integrated to exchange information, shareresources, and collaborate on the clouds.

The proposed IoT-based vehicular data cloud platform sup-ports three new cloud services as indicated in Table I [30].

In this proposed layered architecture, different layers havedifferent purposes. In general, the layers on the bottom provide a

foundational support for the layers on the top. SOA will beapplied to integrate different information and communicationservices and connect in-vehicle and out-vehicle applicationsseamlessly through the vehicular data clouds. SOA allowsvehicular application developers to organize, aggregate, andpackage applications into new business applications services.As a mature technology for enterprise application integration,SOA provides guidelines to integrate heterogeneous web ser-vices, applications, and different middleware systems. Middle-ware is used to hide the implementation details of underliningtechnologies and provides support for the integration of specificapplications deployed on the vehicular data cloud [24]. Byleveraging the SOA-based and IoT-based vehicular data cloudplatform, innovative services can be developed by car manu-facturers, government agencies, and third-party service provides.In this section, we propose two innovative vehicular data cloudservices.

IV. VEHICULAR DATA CLOUD SERVICES

In this section, we are interested in introducing two vehiculardata cloud services as examples of PAAS shown in Section III.

Fig. 1. Architecture for IoT-based vehicular data clouds.

TABLE INOVEL SERVICES FOR IOT-BASED VEHICULAR DATA CLOUDS

HE et al.: DEVELOPING VEHICULAR DATA CLOUD SERVICES IN THE IoT ENVIRONMENT 1589

One service is intelligent parking cloud service and the other oneis the mining vehicular maintenance data service.

A. Intelligent Parking Cloud Service

Finding available parking space is challenging in many citiesand often leads to issues such as congestion, road accidents, andpsychological frustration. To make it easier to find availableparking space, an intelligent parking cloud service that collectsand analyzes geographic location information, parking availabil-ity information, parking space reservation and order information,traffic information and vehicle information through sensordetection and the clouds is needed. Using a modular approach,a software architecture [31] for implementing the intelligentparking cloud service is proposed in Fig. 2.

Each vehicle is pre-enlisted with a transceiver with shorttransmission range (about 1 m) and a processor with simplecomputing capacity. The transceiver can be common devices,such as zigbee, bluetooth devices, and infrared devices, with lowcost. Both the processor and the wireless transceiver are enlistedinto an event data recorder (EDR).Wedesigned a parking lotwithWIFI network, infrared devices, and parking belts to detectmisparked cars. When a car enters the parking lot and heads tothe reserved parking slot, the entrance booth will validate thereservation. If the parking spot is validated, a direction-relatedguidancewill be uploaded to the car for finding the reserved spot.The infrared device, lights, and parking belt will work together todetect and prevent misparking. As shown in Fig. 3, the bluetoothcommunicationwill be activatedwhen the frontwheel presses thebelt-a. The tamper-resistant device (TRD) and belt-a in Fig. 3willvalidate reservation confirmation as necessary. We use an infra-red device to validate whether the car is parked instead of usingthe slot for a temporary purpose. If the parking slot is correctlyparked with cars, the light will show green color; otherwise, thelight shows red color. These sensors connect to computer centerto report the status of every parking slot on an ongoing basis.

We also designed an infrastructure to publish advertisementfrom the parking lot (see Fig. 4). There are wireless transceivertowers in the parking lot andmultiple transceivers (shown as to

) installed on the roadside. The wireless tower in theparking lot can obtain vacant slot information from the computercenter where the status of the parking lot is constantlymonitored.Therefore, wireless tower can broadcast the parking lot informa-tion and parking plan as business strategies for economicalbenefits. As the wireless transmission range is limited, we haveroadside transceivers to relay the parking information to remoteareas, as shown in Fig. 4.

B. Intelligent Parking Service Models

In this section, the parking process has been modeled as abirth–death stochastic process. The parking revenue could bepredicted by using such a model. The birth and the death ofparkingmean that a vehicle enters and exits a parking slot at time, respectively.We were able to obtain the birth and death rate byusing traffic detectors or other sensors [31].We assume that thereis a huge number of parking slots and this number can beconsidered infinite for practical purposes. Let be thenumber of slots in occupied time . We write { } asa birth and death process. Therefore, we can read for all > ,> and . The occupied slot at time is

where is the birth rate and is the death rate. We note theprobability of a car parking event that occurs in ( ) isindependent from the number of occupied parking slots at time .

Fig. 2. Software architecture for intelligent parking cloud service. Fig. 3. Vacancy detections by sensors.

Fig. 4. Parking cloud service.

1590 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

If refers to the probability when the number of parking carsat time is , we can derive the probability of

Based on , we can compute the mean value

C. Mining Vehicular Maintenance Data Service

Another application that we are interested in the proposedvehicular data cloud is tominemaintenance data.Maintenance ofvehicles are frustrated and heavily loaded for drivers. Moreimportantly, auto-manufacturers and auto-parts designer andmanufacturers also desperately seek feedback from end usersto improve the quality of producers and enhance competitioncapability to the foreign auto-makers. We can merge mainte-nance data from both users and repair people.

The merged data are nature language text descriptions ofmaintenance. We place these texts into main files. To dig outauto-parts warranty information, we adopt nature language textmining technologies to these merged texts. As an emergingresearch area, currently limited studies have been publishedregarding how data mining techniques could be applied tovehicular networks or clouds. Few models were developed andtested for mining vehicular data collected from vehicular net-works or data clouds. In this section, we present amodifiedNaïveBayes model [32] and a Logistic Regression model which havebeen adopted in our research of implementing vehicular datamining cloud service which is introduced in Section IV-C1.

Given a document space where all docu-ments are represented in this space.We use for . A set offixed classes is marked as . We use for

and a training set of classi-fied documents. Each labeled document . Aclassifier is used tomap documents to classes: . In thissection, we present an optimized model called Naïve BayesClassifier.

1) Naïve Bayes Classifier: In the Naïve Bayes model, wecalculated the probability of a document locating in the classby using Bayes rules

A document is a list of words: . Thus

The joint probability in

which we assume , i.e., is independentof each other words when . In reality, this assumption is

not always true but this assumption can greatly simplify theproblem and reach acceptable accuracy in real problems.

Therefore, we write

We made efforts to find the “best” class of Naïve Bayesclassification. The best class is typically the maximum of aposteriori (MAP) class

But we usually compute, in practice, by taking the log

Each conditional parameter is used as a weight toindicate the usefulness of an indicator to the class . The priorlog is also used as a weight to indicate the relative frequencyof . The sum of the term weights and the log prior is used tomeasure the amount of available evidence for the document thatis located in the class .We are going to pick the class that has thelargest summation.

We obtain and from the training data set

where is the number of docs in class and is the totalnumber of documents. The conditional probabilities can becalculated as

where refers to the number of words in the trainingdocuments from the class and means the total numberof tokens in training document.

2) Optimization: In this section, we are interested inthe optimization problem of Naïve Bayes Classifier. Leteach document be represented by a word count vector

. We assume for each class , the probabilitydistribution of a document follows the multinomial distributionwith parameter :

The log likelihood is

HE et al.: DEVELOPING VEHICULAR DATA CLOUD SERVICES IN THE IoT ENVIRONMENT 1591

We also assume that the multinomial distribution assumesconditional independence of feature dimensions , giventhe class .Given a training set . Our task,in this step, is to find the best parameters

. Therefore, we translate the model as

. According to the maximum-likelihood

estimation (MLE), the maximum of the joint ( ) likelihood ofthe training set

Therefore, is a constrained optimization problem. We

write is the number of classes and

. It is easy to solve it using

Lagrange multipliers [33] and arrives at

and

The above results are intuitively explained as follows: they areclass frequency in the training data set and the word frequency ofeach class.

3) Logistic Regression Model: Logistic regression model isalso used for both classifying and clustering [34]. As a generativemodel, Naïve Bayes can model the joint , with theassumption of independence on the words of . The Logisticregression, a discriminative model, estimates directly.

We also put an emphasis on the learning classifiersfor a set of training data sets . For a textdocument, the vector consists of transformedword frequencies , from the training document. Thevalues serve as class labels that encode the mem-bership ( ) or nonmembership ( ) of the vector in thecategory of . To map to real number, we calculate the innerproduct between and the parameter vector

Specifically, we calculated the conditional probability inlogistic link function form

Let there are classes . For a class, theprobability of the class can be defined

Therefore, we are looking for that minimizes the followingexpression:

The optimization value of the log likelihood loss form cantypically be solved by Newton–Raphson iterations (or by itera-tive reweighted least squares for logistic regression) [34], [35].

Recall that both Naïve Bayes and Logistic Regression arelinear classifiers. They both divide the documents with ahyperplane. But they differ from each other: Naïve Bayesoptimizes a generative objective function, while LogisticRegression optimizes a discriminative objective function. Inpractice, logistic regression often has higher accuracy whentraining data set is large and on the other hand Naïve Bayes hasan advantage when the size of the training data set is small.

D. Vehicular Data Mining Cloud Service

As vehicular data clouds contain a variety of heterogeneousdata and information resources, effective data mining servicemust be developed to quickly detect dangerous road situations,issue early warning messages, and assist drivers to make in-formed decisions to prevent accidents [36]. Data mining servicescan also be used to assess drivers’ behavior or performance ofvehicles tofind problems in advance. The core of any dataminingservice is the data mining models [37], [38]. So far, few modelswere developed and tested for mining vehicular data collectedfrom vehicular networks or data clouds.

Below is a specialized data mining service for car warrantyearly-warning analysis. We applied the models that we devel-oped in Section IV-C to design and develop the data miningservice. In vehicle manufacturing process, sometimes, somequality issues can be hidden for a long time without beingidentified. Due to a lack of events or signals to correlate severaldiscrete issues, potential problems may not be investigated at all.To avoid accidents, it is important to develop new techniques thatreveal these hidden problems in advance. By using the twomodified data mining models (Naive Bayes Classifier andLogistic Regression Classifier) to cluster and classify the realcar warranty and maintenance data we collected from a localautomobile company, we demonstrated how data mining cloudservice could be used to identify potential issues that couldbecome a problem later. This experiment assumes a new productthat is under development and has some potential but unknownissues. As a result of applying the two data mining models, wewere able to acquire some preliminary results (see Fig. 5). Wefound that the precision in column cross dropped dramatically.The drop is associated with the ’s value which is 3. In otherwords, the model found that there are three groups among all the

1592 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

corpus of natural language text. We show the group distributionin Fig. 6.

V. CHALLENGES AND DIRECTIONS FOR FUTURE WORK

IoT-based vehicular data clouds must be efficient, scalable,secure, and reliable before they could be deployed at a largescale. Existing algorithms and mechanisms are unsatisfactory tomeet all these requirements at the same time. Below is adescription of some of these challenges.

1) Scalability and technology integration: The effectivenessof a vehicular cloud depends on its scalability to handle adynamically changing number of vehicles. In addition tohandling regular traffic, vehicular clouds must be able tohandle traffic spikes or sudden demands caused by specialevents or situations, such as sport games or emergencies.More development on optimization algorithms that coor-dinate virtual machines, storage space, and network band-width to balance server workload and improve computingresource utilization on the vehicular clouds is needed [39].As new devices and technologies are coming out each year,developing effective IoT Middleware that supports inte-gration of these new technologies and devices [40] withexisting in-vehicle technologies from automobile manu-facturers will be a challenge.

2) Performance, reliability and quality of service: As vehi-cles are often on the move, the vehicular networkingand communication is often intermittent or unreliable.More new mechanisms are needed to enhance the

communication reliability with reduced traffic overhead.For example, Chen et al. [41] developed a new transmis-sion protocol to make the conventional Zigbee protocolmore reliable. Cross layer data synchronization mechan-isms should also be designed to minimize the trafficoverhead between layers. Acceleration data compressionalgorithms for resource-constrained sensors, actuators, andother Internet-access devices need improvement to be ableto effectively and efficiently compress a large amount ofraw data generated. Multiple processing units and supportin different cloud data centers are needed to minimize theservice response time, improve availability and stability ofthe service, and increase cloud reliability and fault toler-ance. Real-time evaluation mechanisms regarding theperformance, reliability, and service quality on vehiculardata clouds will have to be further developed.

3) Security and privacy: There are some security and privacyconcerns with vehicular data clouds due to a lack ofestablished infrastructure for authentication and authori-zation [39]. A low security level of vehicular data clouds isunacceptable for vehicular services regarding transporta-tion safety. For example, roadside attackers may mali-ciously send many fake requests to the parking cloudservice and reserve many parking spaces. They can alsosend misleading parking availability information or wronglocation information to the parking cloud service to causechaos. Trust relationships are hard to be built in vehicularclouds because of the large and dynamically changingnumber of vehicles on the road. Balanced security mea-sures are needed to enhance the security and trust of cloudservices without limiting the flexibility of the system. Inaddition, many drivers do not want their vehicle locationsto be tracked or monitored due to the worries about theirprivacy. Reasonable efforts in technology [42], law, andregulation are needed to secure the vehicular data cloudsand prevent unauthorized access to or disclosure of theprivacy data. For example, implementing security authen-tication in vehicular data clouds is required securitycountermeasure.

4) Lack of global standards for device and service integra-tion, security, privacy, architecture, and communications:Global standards are essential to avoid conflicts betweenlocally developed vehicular data clouds [27]. However, asthere are a number of stakeholders involved in vehiculardata clouds, and complex dependencies among thesestakeholders also exist, it is challenging to establish globalstandards to lower the complexity and make vehicular dataclouds more compatible and cost effective. Further effortson standardization are needed to coordinate various effortsand resources for implementing vehicular data clouds.

VI. CONCLUSION

In this paper, we present a novel modular and multilayeredvehicular data cloud platform based on cloud computing and IoTtechnologies. We also discuss how cloud services could bedeveloped to make the vehicular data clouds useful. This studymakes contributions by proposing a novel software architecture

Fig. 5. Clustering results for unknown warranty issues.

Fig. 6. Clusters of unknown warranty issues.

HE et al.: DEVELOPING VEHICULAR DATA CLOUD SERVICES IN THE IoT ENVIRONMENT 1593

for the vehicular data clouds in the IoT environment, which hasthe capabilities to integrate numerous devices available withinvehicles and devices in the road infrastructure.

IoT-based vehicular data clouds are expected to be the back-bone of future ITSswith the ultimate goal ofmaking driving saferand more enjoyable. However, research on integrating IoT withthe vehicular data clouds is still in its infancy and existing studyon this topic is highly insufficient. Tomake vehicular data cloudsuseful, numerous services, such as road navigation, trafficmanagement, remote monitoring, urban surveillance, informa-tion and entertainment, and business intelligence [43]–[47], needto be developed and deployed on vehicular data clouds. Anumber of challenges such as security, privacy, scalability,reliability, quality of service, and lack of global standards stillexist. Due to the complexity involved in implementing vehicularclouds and integrating various devices and systems with vehic-ular clouds [48]–[51], a systematic approach and collaborationamong academia, the automobile companies, law enforcement,government authorities, standardization groups, and cloud ser-vice providers are needed to address these challenges. Thoughwith many challenges, IoT and cloud computing providetremendous opportunities for technology innovation in the auto-mobile industry [52], [53] and will serve as enabling infrastruc-tures for developing vehicular data clouds.

REFERENCES

[1] A. Iwai and M. Aoyama, “Automotive cloud service systems based onservice-oriented architecture and its evaluation,” in Proc. IEEE Int. Conf.Cloud Comput., Washington, DC, USA, 2011, pp. 638–645.

[2] Y. Zhao, “Telematics: Safe and fun driving,” IEEE Intell. Syst., vol. 17,no. 1, pp. 10–14, Jan./Feb. 2002.

[3] S. Bitam and A. Mellouk, “ITS-cloud: Cloud computing for intelli-gent transportation system,” in Proc. IEEE Global Commun. Conf.,Anaheim, CA, USA, 2012, pp. 2054–2059.

[4] P. Jaworski, T. Edwards, J. Moore, and K. Burnham, “Cloud computingconcept for intelligent transportation systems,” inProc. 14th Int. IEEEConf.Intell. Transp. Syst., Washington, DC, USA, 2011, pp. 391–936.

[5] Z. Pang, L. Zheng, J. Tian, S. Kao-Walter, E. Dubrova, and Q. Chen,“Design of a terminal solution for integration of in-home health caredevices and services towards the internet-of-things,” Enterprise Inf.Syst., to be published, 2014.

[6] L. Xu, “Enterprise systems: State-of-the-art and future trends,” IEEE Trans.Ind. Informat., vol. 7, no. 4, pp. 630–640, Nov. 2011.

[7] S. Li, L. Xu, andX.Wang, “Compressed sensing signal and data acquisitionin wireless sensor networks and internet of things,” IEEE Trans. Ind.Informat., vol. 9, no. 4, pp. 2177–2186, Nov. 2013.

[8] Y. Li, M. Hou, H. Liu, and Y. Liu, “Towards a theoretical framework ofstrategic decision, supporting capability and information sharing under thecontext of internet of things,” Inf. Technol. Manag., vol. 13, no. 4,pp. 205–216, 2012.

[9] P. Papadimitratos, A. La Fortelle, K. Evenssen, R. Brignolo, andS. Cosenza, “Vehicular communication systems: Enabling technologies,applications, and future outlook on intelligent transportation,” IEEE Com-mun. Mag., vol. 47, no. 11, pp. 84–95, Nov. 2009.

[10] H.Abid,L.T.T.Phuong,J.Wang,S.Lee,andS.Qaisar,“V-Cloud:Vehicularcyber-physical systems and cloud computing,” in Proc. 4th Int. Symp.Appl. Sci. Biomed. Commun. Technol., Barcelona, Spain, 2011, p. 165.

[11] S. Olariu, I. Khalil, andM.Abuelela, “TakingVANET to the clouds,” Int. J.Pervasive Comput. Commun., vol. 7, no. 1, pp. 7–21, 2011.

[12] Y. Leng and L. Zhao, “Novel design of intelligent internet-of-vehiclesmanagement system based on cloud-computing and internet-of-things,”inProc. Int. Conf. Electron.Mech. Eng. Inf. Technol., Harbin,Heilongjiang,China, 2011, vol. 6, pp. 3190–3193.

[13] R. Hussain, J. Son, H. Eun, S. Kim, and H. Oh, “Rethinking vehicularcommunications: Merging VANET with cloud computing,” in Proc. IEEE4th Int. Conf. Cloud Comput. Technol. Sci., Taipei, Taiwan, 2012,pp. 606–609.

[14] R. Mietzner, F. Leymann, and T. Unger, “Horizontal and vertical combi-nation ofmulti-tenancy patterns in service-oriented application,”EnterpriseInf. Syst., vol. 5, no. 1, pp. 59–77, 2011.

[15] H. Panetto and J. Cecil, “Information systems for enterprise integration,interoperability and networking: Theory and applications,” Enterprise Inf.Syst., vol. 7, no. 1, pp. 1–6, 2013.

[16] S. Hachani, L. Gzara, and H. Verjus, “A service-oriented approach forflexible process support within enterprises: Application on PLM systems,”Enterprise Inf. Syst., vol. 7, no. 1, pp. 79–99, 2013.

[17] F. Tao, H. Guo, L. Zhang, and Y. Cheng, “Modelling of combinablerelationship-based composition service network and the theoretical proofof its scale-free characteristics,” Enterprise Inf. Syst., vol. 6, no. 4,pp. 373–404, 2012.

[18] D. Paulraj, S. Swamynathan, and M. Madhaiyan, “Process model-basedatomic service discovery and composition of composite semantic webservices using web ontology language for services (OWL-S),” EnterpriseInf. Syst., vol. 6, no. 4, pp. 445–471, 2012.

[19] J. Wang, J. Cho, S. Lee, and T. Ma, “Real time services for futurecloud computing enabled vehicle networks,” in Proc. Int. Conf. WirelessCommun. Signal Process., Nanjing, China, 2011, pp. 1–5.

[20] Y. Qin, D. Huang, and X. Zhang, “VehiCloud: Cloud computingfacilitating routing in vehicular networks,” in Proc. IEEE 11th Int.Conf. Trust Secur. Privacy Comput. Commun., Liverpool, U.K., 2012,pp. 1438–1445.

[21] H.Wang,W. He, and F. K.Wang, “Enterprise cloud service architectures,”Inf. Technol. Manag., vol. 13, no. 4, pp. 445–454, 2012.

[22] Q. Li, Z. Y. Wang, W. H. Li, J. Li, C. Wang, and R. Y. Du, “Applicationsintegration in a hybrid cloud computing environment: Modeling andplatform,” Enterprise Inf. Syst., vol. 7, no. 3, pp. 237–271, 2012.

[23] L. Ren, L. Zhang, F. Tao, X. Zhang, Y. L. Luo, and Y. Zhang,“A methodology towards virtualization-based high performance simula-tion platform supporting multidisciplinary design of complex products,”Enterprise Inf. Syst., vol. 6, no. 3, pp. 267–290, 2012.

[24] European Commission Information Society, Internet of Things in 2020:A Roadmap for the Future [Online]. Available: www.iot-visitthefuture.eu.

[25] L. Atzori, A. Iera, and G. Morabito, “The internet of things: A survey,”Comput. Netw., vol. 54, no. 15, pp. 2787–2805, 2010.

[26] C. Speed and D. Shingleton, “An internet of cars: Connecting the flow ofthings to people, artefacts, environments and businesses,” inProc. 6th ACMWorkshop Next Gener. Mobile Comput. Dynam. Personalised TravelPlann. ACM, Ambleside, U.K., 2012, pp. 11–12.

[27] E. Qin, Y. Long, C. Zhang, and L. Huang, “Cloud computing and theinternet of things: Technology innovation in automobile service,” inHumanInterface and the Management of Information. Information and Interactionfor Health, Safety, Mobility and Complex Environments. Berlin, Germany:Springer-Verlag, 2013, pp. 173–180.

[28] W. Lumpkins, “The internet of things meets cloud computing,” IEEEConsum. Electron. Mag., vol. 2, no. 2, pp. 47–51, Apr. 2013.

[29] Y. Zhang, B. Chen, and X. Lu, “Intelligent monitoring system on refrigera-tor trucks based on the internet of things,” Wireless Communications andApplications. Berlin, Germany: Springer-Verlag, 2012, pp. 201–206.

[30] S. Arif, S. Olariu, J. Wang, G. Yan, W. Yang, and I. Khalil, “Datacenter atthe airport: Reasoning about time-dependent parking lot occupancy,”IEEE Trans. Parallel Distrib. Syst., vol. 23, no. 11, pp. 2067–2080,Nov. 2012.

[31] G. Yan,W. Yang, D. B. Rawat, and S. Olariu, “SmartParking: A secure andintelligent parking system,” IEEE Intell. Transp. Syst. Mag., vol. 3, no. 1,pp. 18–30, Apr. 2011.

[32] D. Lowd and P. Domingos, “Naive Bayes models for probability estima-tion,” in Proc. 22nd Int. Conf. Mach. Learn., Bonn, Germany: ACM, 2005,pp. 529–536.

[33] R. T. Rockafellar, “Lagrange multipliers and optimality,” SIAM Rev.,vol. 35, no. 2, pp. 183–238, 1993.

[34] G. Qian, Y. Wu, and Q. Shao, “A procedure for estimating the number ofclusters in logistic regression clustering,” J. Classification, vol. 26, no. 2,pp. 183–199, 2009.

[35] T. J. Hastie and D. Pregibon, Generalized Linear Models. Belmont, CA,USA: Wadsworth & Brooks/Cole, 1992.

[36] M. Faezipour, M. Nourani, A. Saeed, and S. Addepalli, “Progress andchallenges in intelligent vehicle area networks,” Commun. ACM, vol. 55,no. 2, pp. 90–100, 2012.

[37] L. Bulysheva and A. Bulyshev, “Segmentation modeling algorithm: Anovel algorithm in data mining,” Inf. Technol. Manag., vol. 13, no. 4,pp. 263–271, 2012.

[38] L. Zeng, L. Li, L. Duan, K. Lu, Z. Shi, M. Wang, W. Wu, and P. Luo,“Distributed data mining: A survey,” Inf. Technol. Manag., vol. 13, no. 4,pp. 403–409, 2012.

1594 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

[39] G. Yan, D. Wen, S. Olariu, and M. C. Weigle, “Security challenges invehicular cloud computing,” IEEE Trans. Intell. Transp. Syst., vol. 14,no. 1, pp. 284–294, Mar. 2013.

[40] S. Li, L. Xu, X. Wang, and J. Wang, “Integration of hybrid wirelessnetworks in cloud services oriented enterprise information systems,”Enterprise Inf. Syst., vol. 6, no. 2, pp. 165–187, 2012.

[41] S. K. Chen, T. Kao, C. T. Chan, C. N. Huang, C. Y. Chiang, C. Y. Lai et al.,“A reliable transmission protocol for ZigBee-based wireless patient moni-toring,” IEEE Trans. Inf. Technol. Biomed., vol. 16, no. 1, pp. 6–16,Jan. 2012.

[42] L. Li, “Technology designed to combat fakes in the global supply chain,”Bus. Horizons, vol. 56, no. 2, pp. 167–177, 2013.

[43] L. Duan and L. D. Xu, “Business intelligence for enterprise systems: Asurvey,” IEEE Trans. Ind. Informat., vol. 8, no. 3, pp. 679–687, Aug. 2012.

[44] M. Gerla, “Vehicular cloud computing,” in Proc. 11th IEEE Annu.Mediterranean Ad Hoc Netw. Workshop, Ayia Napa, Cyprus, 2012,pp. 152–155.

[45] S. Fang, L. Xu, H. Pei, Y. Liu, Z. Liu, Y. Zhu et al., “An integrated approachto snowmelt flood forecasting in water resourcemanagement,” IEEE Trans.Ind. Informat., vol. 10, no. 1, pp. 548–558, Feb. 2013.

[46] L. Zeng, L. Li, and L. Duan, “Business intelligence in enterprise computingenvironment,” Inf. Technol. Manag., vol. 13, no. 4, pp. 297–310, 2012.

[47] J. E. Ingvaldsen and J. A. Gulla, “Industrial application of semantic processmining,” Enterprise Inf. Syst., vol. 6, no. 2, pp. 139–163, 2012.

[48] Y. Xing, L. Li, and Z. Bi, “Operations research (OR) in service industries: Acomprehensive review,” Syst. Res. Behav. Sci., vol. 30, pp. 300–353, 2013.

[49] J. Wan and J. D. Jones, “Managing IT service management implementationcomplexity: From the perspective of the Warfield version of systemsscience,” Enterprise Inf. Syst., vol. 7, no. 4, pp. 490–522, 2011.

[50] S. Wang, L. Li, K. Wang, and J. D. Jones, “e-business systems integration:A systems perspective,” Inf. Technol. Manag., vol. 13, no. 4,pp. 233–249, 2012.

[51] L. D. Xu, “Introduction: Systems science in industrial sectors,” Syst. Res.Behav. Sci., vol. 30, no. 3, pp. 211–213, 2013.

[52] G. Goggin, “Driving the internet: Mobile internets, cars, and the social,”Future Internet, vol. 4, no. 1, pp. 306–321, 2012.

[53] J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, “Internet of things(IoT): A vision, architectural elements, and future directions,” FutureGener. Comput. Syst., vol. 29, no. 7, pp. 1645–1660, 2013.

Wu He received the B.S. degree in computer sciencefromDongHua University, Shanghai, China, in 1998,and the Ph.D. degree in information science from theUniversity of Missouri, Columbia, MO, USA, in2006.His research interests include enterprise applica-

tions, data mining, cyber security, and knowledgemanagement.

Gongjun Yan received the Ph.D. degree in computerscience fromOld Dominion University, Norfolk, VA,USA, in 2010.Currently, he is an Assistant Professor of Computer

Science with the University of Southern Indiana,Evansville, IN, USA. He appliedmathematics modelsand computer simulations to analyze research pro-blems in information security, vehicular ad hoc net-works, and wireless communications.

Li Da Xu (M’86–SM’11) received theM.S. degree ininformation science and engineering from the Univer-sity of Science and Technology of China, Hefei,China, in 1981, and the Ph.D. degree in systemsscience and engineering from Portland State Univer-sity, Portland, OR, USA, in 1986.Dr. Xu serves as the Founding Chair of IFIP TC8

WG8.9 and the Founding Chair of the IEEE SMCSociety Technical Committee on Enterprise Informa-tion Systems.

HE et al.: DEVELOPING VEHICULAR DATA CLOUD SERVICES IN THE IoT ENVIRONMENT 1595