资源描述:
《Human Tracking by Adaptive Kalman Filtering and Multiple Kernels Tracking with Projected Gradients.pdf》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、HumanTrackingbyAdaptiveKalmanFilteringandMultipleKernelsTrackingwithProjectedGradientsChun-TeChu,Jenq-NengHwangShen-ZhengWang,Yi-YuanChenDepartmentofElectricalEngineering,Box352500ServiceSystemsTechnologyCenter,UniversityofWashingtonIndustrialTechnologyResearchInstituteSeattle,WA98195,USAHs
2、inchu,Taiwan31040,R.O.C{ctchu,hwang}@u.washington.edu{st,yiyuan}@itri.org.twAbstract—Kernelbasedtrackershavebeenproventobeaoccludedbyothers,theinformationforthetargetreducespromisingapproachinvideoobjecttracking.Theuseofsinglegreatly,andthetrackertendstolosetheobject.Inthissituation,kernelo
3、ftensuffersfromocclusionsincethevisualinformationissinglekerneltrackingisnotreliableforlocatingtheobject.notsufficientforkernelusage.Hence,multipleinter-relatedHence,multiplekernelstrackingshouldbeemployed.Wekernelshavebeenutilizedfortrackingincomplicatedscenarios.presentaprojectedgradientb
4、asedmultiplekernelstrackingThispaperembedsthemultiplekernelstrackingintoaKalmanschemetoeffectivelytrackapre-identifiedtargetinheavyfiltering-basedtrackingsystem,whichusesKalmanpredictionascrowd[2].Besidestheindividualtrackingwithineachkernel,theinitialpositionforthemultiplekernelstracking,a
5、ndapplieskernelsarecontrolledbysomepredefinedconstraintssuchastheresultofthelatterasthemeasurementtotheKalmanupdate.thegeometricalrelationshipbetweenthem.Moreover,inthisThestatetransitionandnoisecovariancematricesusedinpaperwefurtherembedthemultiplekernelstrackingwithaKalmanfilterarealsodyn
6、amicallyupdatedbytheoutputofKalmanfiltertrackingsystemtoallowtrackingofallmovingmultiplekernelstracking.Severalsimulationresultshavebeenobjectssimultaneously.TheKalmanpredictiongivestheinitialdonetoshowtherobustnessoftheproposedsystemwhichcansuccessfullytrackallthevideoobjectsunderocclusion
7、.positionofthemultiplekernelstracking,andtheresultofthelatterwillbethemeasurementfortheKalmanupdate.TheKeywords-Kalmanfilters,kernel-basedtracking,meanshiftpropercouplingofthetwomakesourtrackingapproachrobusttotheocclusionproblem.I.INTRODUCTIONAfullyauto