QueryingTrajectoriesUsingFlexiblePatterns
UniversityofCaliforniaRiverside,CA92507,USA
MarcosR.Vieira
mvieira@cs.ucr.edu
ESRI
Redlands,CA92373,USA
PetkoBakalov
pbakalov@esri.com
UniversityofCaliforniaRiverside,CA92507,USA
VassilisJ.Tsotras
tsotras@cs.ucr.edu
ABSTRACT
ThewideadaptationofGPSandcellulartechnologieshascreatedmanyapplicationsthatcollectandmaintainlargerepositoriesofdataintheformoftrajectories.Previousworkonquerying/analyzingtrajectorialdatatypicallyfallsintomethodsthateitheraddressspatialrangeandNNqueries,or,similaritybasedqueries.Nevertheless,trajectoriesarecomplexobjectswhosebehaviorovertimeandspacecanbebettercapturedasasequenceofinterestingevents.Wethusfacilitatetheuseofmotion“pattern”querieswhichal-lowtheusertoselecttrajectoriesbasedonspeci cmotionpatterns.Suchpatternsaredescribedasregularexpressionsoveraspatialalphabetthatcanbeimplicitlyorexplicitlyanchoredtothetimedomain.Moreover,weareinterestedin“ exible”patternsthatallowtheusertoinclude“variables”inthequerypatternandthusgreatlyincreaseitsexpressivepower.Inthispaperweintroduceaframeworkfore cientprocessingof exiblepatternqueries.Theframeworkin-cludesanunderlyingindexingstructureandalgorithmsforqueryprocessingusingdi erentevaluationstrategies.Anextensiveperformanceevaluationofthisframeworkshowssigni cantperformanceimprovementwhencomparedtoex-istingsolutions.
ingobjectsbasedontheircurrentpositions(e.g.[12][17]),and,(b)queryingtrajectoryarchives,whichisalsothefocusofthispaper.Recentresearche ortsonqueryingtrajectoryarchiveshasconcentratedon(i)traditionalspatiotemporalqueries,suchasrangesearchesandnearestneighbors(e.g.[10][18])(e.g. ndingalltrajectoriesthatpassedbydown-townLAat10:30am),or(ii)similarity/clusteringbasedtasks(e.g.[13][14]),suchasextractingsimilarmovementpatternsandperiodicitiesfromatrajectoryarchive(e.g. ndingalltrajectoriesinthearchivethataresimilartoagivenquerytrajectoryaccordingtosomesimilaritymea-sure).
However,giventhenatureoftrajectoriesastypicallylongsequencesofevents,asinglerangepredicatemayprovidetoomanyresults(manytrajectoriespassedthroughdown-townLA)whileasimilarity-basedquerymaybetoorestric-tive(notmanytrajectoriesmatchthefullextentorlargepartofthequerytrajectory).Wethusadvocateadi erentapproach,i.e.,usingmotion“pattern”queries.Amotionpatternqueryspeci esacombinationofpredicatesthatcanthuscaptureonlythepartsofthetrajectoriesthatareofinteresttotheuser.Forexample:“ ndalltrajectoriesthat rstwentbydowntownLA,laterbyWestHollywoodandendedupinBeverlyHills”.Thisquerysimplyprovidesacollectionofrangepredicatesthathavetobesatis edinthespeci edorder.OnecanalsoaddNearest-Neighbor(NN)conditionsaswell(intheabovequery:“...andtheywereclosesttotheLAXairport”)aswellasexplicittimecon-straints:“endedupinBeverlyHillsat10am”.Concep-tually,motionpatternqueriescoverthequerychoicesbe-tweentheabovetwoextremes(singlepredicatesandsimi-larityqueries).
Inthispaperweintroduceageneralandpowerfulframe-workthatdescribespatternqueriesasregularexpressionsovera nitespatialalphabet.Eachletterinthespatialal-phabetcorrespondstoanon-overlappingregion;theirunioncoversthewholespacewherethetrajectorieslie.Wenotethattherearevariousadvantagesfromthesechoices.(1)Theuseofnon-overlappingregionsisnatural:trajectoriescorrespondtorealentitiesandhenceatrajectorycanbeinasingleregionatagiventime.(2)Rawtrajectorydatatypicallycomefromsensors,GPSand/orRFIDreaders,etc.andprovideextradetailthatbecomescumbersometoquery.Instead,theregionso eramoreuser-friendlywaytoexpressqueriessincetheuserismorefamiliarwiththespatialre-gions[7](forexample,Downtown,LAX,etc.)(3)Theuse
1.INTRODUCTION
Thewideavailabilityoflocationandmobiletechnologies(cheapGPSdevices,ubiquitouscellularnetworks)aswellastheimprovedlocationaccuracy(A-GPS)hasenabledmanyapplicationsthatgenerateandmaintaindataintheformoftrajectories.ExamplesincludeAccuTracking,tracNET24,PathIntelligence’sFootPath,InSTEDD’sGeoChat,amongmanyothers.Eachtrajectoryhasauniqueidenti erandconsistsoflocationdatagatheredforaspeci cmovingob-jectoveranorderedsequenceoftimeinstants.Giventhehighdatavolume,moree cienttechniquesforqueryeval-uationovertrajectorydataareneeded.
Previousworkonqueryingtrajectoriescanbedividedintwocategories,(a)queryingthefuturemovementsofmov-
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforpro torcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthe rstpage.Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspeci cpermissionand/orafee.
EDBT2010,March22–26,2010,Lausanne,Switzerland.Copyright2010ACM978-1-60558-945-9/10/0003...$10.00
Modeling of Moving Objects and Querying Videos by Trajectories Content-based retrieval of videos in databases is a technique which has attracted considerable...
Modeling of Moving Objects and Querying Videos by Trajectories Content-based retrieval of videos in databases is a technique which has attracted considerable...
热门文档