基于Kalman滤波的行人跟踪方法研究

96200912JournalofTransportationSystemsEngineeringandInformationTechnologyVol19No16December2009:100926744(2009)0620148206Kalman,3,,(,100044):,,,,.Kalman.,,HSV.,,.,Kalman,,,.,,,.:;;;Kalman:U491:AStudyonPedestrianTrackingBasedonKalmanFilterLIJuan,SHAOChun2fu,YANGLi2ya,LIQi(MOEKeyLaboratoryforTransportationComplexSystemsTheoryandTechnology,BeijingJiaotongUniversity,Beijing100044,China)Abstract:PedestriantrackingisanimportantpartofIntelligentTransportationSystems(ITS)application.Anewrobustsystemforpedestriantrackingbasedoncomputervisionisproposedinthispaper.First,improvedGaussianmixturemodelisusedtodetectpedestrian.Togetbettersegmentationresult,morphologicalreconstructionisuti2lizedtoremoveshadows.Then,pedestriantrackingisachievedbyfeaturefusionandpredictionmethodology.ThetrajectoriesareincorporatedbytheKalmanfiltertodeterminethesearchwindows,whichareusedtoestablishthematchingmatrixattwosuccessiveframesbycomparingwithdetectedpedestrians.Finally,anobjectassociationstrategyisincorporatedtodealwithobjecttrackingincaseofmergingandsplittingbythetemplateintegratedspa2tialpositionandshapeinformation.Thetrackingmethodistestedunderrealscenarios.Elaborateexperimentre2sultsshowgoodrobustnessandhighefficiencyofthismethod.Keywords:intelligenttransportationsystems(ITS);pedestriantracking;Gaussianmixturemodel;KalmanfilterCLCnumber:U491Documentcode:A:2009201222:2009203224:2009203231:(973)(2006CB705500);(50778015);(07XND012).:(1980-),,,.3:[e-mail]1,.,,[1-5].,.,,[6],,,.,,.4:[7].[8],,.[9],.[10],,,.[11],,,.,,.,,.,HSV.,.Kalman,.,.,,.2,..2.1.,,.K,[12].,,,.XtP(Xt)=Kk=1k,t(Xt,k,t,2k,t)(1)K;k,ttk;k,t2ktk;(Xt,k,t,2k,t)tk,(Xt,k,t,2k,t)=1(2)n2k,te-12(Xt-k,t)T-2k,t(Xt-k,t)(2),.EM(),,.2.2F1.[13],HSV.HSV,H,S,V,,RGB[14].HSV,F2.,.,[15]:F=F3(F2ÝB)(3)F39416Kalman;B,.F,1.1Fig.1Foregroundimagesaftershadowremoval3,,,..(px,py)(vx,vy).2,(px,py).(sx,sy)t.vx=sxt,vy=syt(4)2Fig.2Boundingboxofapedestrianandcorrespondingcentroid(W)(L)(A)(R)(C),:R=WL(5)C=4AP2(6)P.,Temp=[px,py,vx,vy,W,L,A,R,C].,.4,.,.,Kalman,,.,.4.1,2,.Kalman,,,[16].xt=Axt-1+But+wt(7)zt=Hxt+vt(8)xttn;zkt.AH.wtvt,.Kalmanxt(px,py,vx,vy),(px,py)(vx,vy).,AA=10t0010t00100001(9)051200912,HH=10000100(10)4.2Kalman,.Kalman,,.,3,,,.3Fig.3Asearchwindowforthecandidatematchedobjectsofanobject,1,,0,.1.45.,,.Ó4(a),,,;Ó1223,,23,4(b);Ó344,,4(c);Ó4(d)5,,;Ó4(e)5,,.1Table1Trackingmatchingmatrix123451100002011003000104000105000004Fig.4Resultofobjectsmatching4.3(1).,,,..,kalman,1516Kalman,.(2).,,,,.,,.,,,,.:S=|A-Atemp|PAtemp+|R-Rtemp|PRtemp+|C-Ctemp|PCtemp(11)AtempRtempCtemp.5,,5(c),5(d),,.5Fig.5Multiplepedestrianstrackingwithocclusion5VC++6.0OpenCV.IntelCore2CPU,1.83GHz,512MB,WinXP.320240,25.6.,.,6(a),6(b).6Fig.6Trajectoryofthepedestrian2,48ms,,.3,,..,,,.2Table2Meanprocessingtime(ms)22818483Table3Trackingresultstatistics(%)411004097.612.43792.537.56Kalman.,HSV.Kalman251200912.,.,,,.:[1]GHZhang,RPAvery,YHWang.Avideo2basedvehi2cledetectionandclassificationsystemforreal2timetrafficdatacollectionusinguncalibratedvideocameras[J].TransportationResearchRecord,No.1993,2007:138-147.[2]HCheng,NNZheng,XTZhang,JJQin,HvandeWetering.interactiveroadsituationanalysisfordriveras2sistanceandsafetywarningsystems:frameworkandalgo2rithms[J].IEEETransactionsonIntelligentTransportationSystems,2007,8(1):157-167.[3],,.[J].,2006,6:66-70.[SHAOCF,ZHAOY,WUG.Reviewofroadtrafficdatacollec2tiontechnology[J].ModernTransportationTechnology,2006,6:66-70.][4]CFShao,YZhao,HYue,XZhang.Studyonavisionbasedsystemforreal2timecollectionofmixedtrafficdata[C].ProceedingsofICTTS2006:606-614.[5],,,.[J].,2008,8(4):23-29.[SHAOCF,LIJ,ZHAOY,etal.Asurveyofpedestriantrafficdatacollectionmethodbasedonvideoimageprocessing[J].JournalofTransportationSystemsEngineeringandInformationTechnology,2008,8(4):23-29.][6],,,.[J].,2007,7(4):47-51.[YUEH,SHAOCF,ZHAOY,etal.Astudyonmovingpedestriantrackingbasedonvid2eosequences[J].JournalofTransportationSystemsEngi2neeringandInformationTechnology,2007,7(4):47-51.][7]BLiu,OJesorsky,RKompe.Robustreal2timemultipleobjecttrackingintrafficscenesusinganopticalmatrixrangesensor[C].IEEEInternationalConferenceonIntel2ligentTransportationSystems,2007:742-747.[8]JepsonAD,FleetDJ,El2MaraghiTF.Robustonlineappearancemodelsforvisualtracking[J].IEEETransac2tionsonPatternAnalysisandMachineIntelligence,2003,25(10):1296-1311.[9]ParagiosN,DericheR.Geodesicactivecontoursandlevelsetsforthedetectionandtrackingofmovingobjects[C].IEEETrans.onPatternAnalysisandMachineIntelli2gence,2000,22(3):266-280.[10]SegenJ,PingaliS.Acamera2basedsystemfortrackingpeopleinrealtime[C].ProcInternationalConferenceonPaternrecognition,Vienna,1996:63-67.[11]RohrK.Towardsmodel2basedrecognitionofhumanmove2mentsinsequences[C].ComputerVision,GraphicsandImageProcessing,1994:94-115.[12]StaufferC,GrimsonWEL.Learningpatternsofactivityusingreal2timetracking[C].IEEETransactionsonPatternAnalysisandMachineIntelligence,2000:747-757.[13],,.[J].,2008,35(3):21-25.[WANGY,TANYH,TIANJW.Videioseg2mentationalgorithmwithGaussianMixtureModelandshadowremoval[J].Opto2ElectronicEnginerring,