siyuan liu *#, yunhuai liu *, lionel m. ni *# +, jianping fan #, minglu li + * hong kong university...
TRANSCRIPT
Siyuan Liu*#, Yunhuai Liu*, Lionel M. Ni*# +, Jianping Fan#, Minglu Li+
*Hong Kong University of Science and Technology#Shenzhen Institutes of Advanced Technology, Chinese Academy of
Sciences+ Shanghai Jiao Tong University
July 27th, 2010@SIGKDD 2010
Towards Mobility-based Clustering
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
Smart City [1]
China's urbanizationMassive issues and problems
City monitoring and managementPervasive information and knowledge
Digital technologyData collection, storage and miningReal life data sets
Vehicle GPS data sets (one year, two cities)Mobile phone networks data sets (one year, two
cities)
Hot spot detection in the city
[1] Smart City Research Group. http://www.cse.ust.hk/scrg
Motivation
Vehicle instant locations of sample taxis at 13:00PM on 12th Dec, 2006
Traffic congesti
on
Event detectio
n
Commercial
promotion
Crowded spots and areas in the
city
Data setIdeal case: we should have all information
of all vehicles in the cityReality: only a sample set of all vehicles
Taxi GPS data (ID, location, speed, time, direction, status)
0.3% of the two million vehicles in Shanghai
Could we utilize such a very limited sample set to detect
hot spot in the city?
Challenges
Extremely limited sample
set
Dense?
Sparse?
Sparse!
Dense!Notable location
error
Could density based clustering handle it?
MethodologyObservation
The low speed may indicate that the area is crowded
MethodMobility-based clusteringStudy the speed (mobility) instead of the
density
Moving objects as sensors
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
Related Work
•Partitioning methods•Hierarchical methods•Density based methods•Grid based methods•Model based methodsetc.
•Raw data based methods•Feature based methods•Model based methodsetc.
What if the mobility is high?What if the density is poor?What if the location is lossy?
10
Density based clustering
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
Mobility-based Clustering
12Roadmap
Object Mobility ModelSpeed estimation
Road network gridInterpolation
Direction distinguishing
Speed spectrum of road direction
Speed spectrum of reverse direction
NanPu Bridge13
Spot Crowdedness ModelLinear crowdedness function
Statistical crowdedness function
( )( )( ) max
max min
( ) ( )( ) ( ) (1 )
( ) ( )l
ttt
l lL Lv l v l
l lv l v l
( )( ) ( )( ) ( ) (1 ) (1 ( ( )))ltt tl lS Sl l v v l
14
Crowdedness Model ValidationValidation
1. Taxis traces 2. Buses tracesO r i g i n a l d a t a s e t Input
set Ф
Test set Фc
Randomly split to two parts
Crowdedness computation
Mobility estimation
ValidationMobility
error
15
Learning in Practice
Characterizing spots
α, г
Sensor object profiling
Hot spots and hot regions
Temporal hot spots
Evolutionary hot regions
Spot crowdednes
s
16
Learning in Mobility Based Clustering
Learning in PracticeCharacterizing spots
17
Learning in PracticeSensor object profiling
18
Learning in PracticeSensor object profiling
19
Learning in PracticeHot spots and hot regions
( ) ( ) ( ) { | 0, ' , ( ', ) ( ') ( )}t t ts sHot spot Ls l l A dist l l l l
( ) ( ) ( ) { | , ( ) }t tth s thHot region Rs l l A l
20
Hot spot, even sparse sample points
NOT hot spot, even dense sample points
Learning in PracticeTemporal hot spots
Event detection Temporal consistence
Learning in PracticeEvolutionary hot regions
Area difference ratioCrowdedness difference ratio
22
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
Field Study Evaluation
24
Field Study Evaluation
25
OutlineIntroductionRelated workMobility based clusteringField study evaluationConclusion
Conclusion and Future WorkContributions
Mobility-based clustering modelKey factors on spot crowdednessHot spots and hot regions
Future workMore accurate speed informationMore accurate location information
Thanks for your attention!