资源描述：
《mean shift Tracking》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、IEEECVPR2000Real-TimeTrackingofNon-RigidObjectsusingMeanShiftDorinComaniciuVisvanathanRameshPeterMeerImaging&VisualizationDepartmentElectrical&ComputerEngineeringDepartmentSiemensCorporateResearchRutgersUniversity755CollegeRoadEast,Princeton,NJ0854094BrettRoad,Piscataway,NJ08855Abstract2MeanShif
2、tAnalysisAnewmethodforreal-timetrackingofnon-rigidob-Wede nenextthesamplemeanshift,introducethejectsseenfromamovingcameraisproposed.Thecen-iterativemeanshiftprocedure,andpresentanewthe-tralcomputationalmoduleisbasedonthemeanshiftoremshowingtheconvergenceforkernelswithconvexiterationsand ndsthemo
3、stprobabletargetpositioninandmonotonicpro les.Forapplicationsofthemeanthecurrentframe.Thedissimilaritybetweenthetargetshiftpropertyinlowlevelvision( ltering,segmenta-model(itscolordistribution)andthetargetcandidatestion)see[6].isexpressedbyametricderivedfromtheBhattacharyya2.1SampleMeanShiftcoe
4、cient.ThetheoreticalanalysisoftheapproachshowsthatitrelatestotheBayesianframeworkwhileGivenasetfxgofnpointsinthed-ii=1:::ndprovidingapractical,fastandecientsolution.ThedimensionalspaceR,themultivariatekerneldensitycapabilityofthetrackertohandleinreal-timepartialestimatewithkernelK(x)andwindowra
5、dius(band-occlusions,signi cantclutter,andtargetscalevaria-width)h,computedinthepointxisgivenbytions,isdemonstratedforseveralimagesequences.nX1x;xi^f(x)=K:(1)d1Introductionnhhi=1TheecienttrackingofvisualfeaturesincomplexTheminimizationoftheaverageglobalerrorbetweenenvironmentsisachallengingta
6、skforthevisioncom-theestimateandthetruedensityyieldsthemultivariatemunity.Real-timeapplicationssuchassurveillanceandEpanechnikovkernel[25,p.139]monitoring[10],perceptualuserinterfaces[4],smart;112c(d+2)(1;kxk)ifkxk<1rooms[16,28],andvideocompression[12]allrequired2K(x)=E0otherwisetheabilitytotra
7、ckmovingobjects.Thecomputationalcomplexityofthetrackeriscriticalformostapplica-(2)wherecisthevolumeoftheunitd-dimensionalsphere.dtions,onlyasmallpercentageofasystemresourcesbe-Anothercommonlyusedkernelisthemultivariatenor-in
