摄像头网络环境下多目标跟踪算法分析-analysis of multi-target tracking algorithm in camera network environment.docxVIP

AbstractSingle-cameravisualtrackingsystemhasmanyunresolvedissues,includingthetargetocclusion,thelimitationofcamerafieldofvision,cannotbeall-roundtrackingandsoon,butmulti-cameratrackingsystemcanwellovercometheseproblems.Thetraditionaltrackingmethodsoftenignoreproblemssuchasinaccurategeometricconstraintsorviolatethecollectionrestrictions,butrequiresmoresophisticatedmethodstodealwiththisproblem.Basedonpreviousstudies,thepaperfocusesonhowtoimprovethemulti-cameramultipletargetdetectionaccuracyandtheproblemsoftargetfusionandtrackingafterdetection,torealizethemultipletargettrackingundermulti-cameraenvironment.Themainworkofthispaperareasfollows:Firstofall,thepaperanalyzestargetdetectionalgorithmbasedonhumanprincipalaxis,accordingtothecharacteristicsthatthehumanbodyprincipalaxisalmostcompletelysymmetricalonbothsides,wedetectthetargetsovermulti-camerasbyapplingtheLeastMedianofSquarestodeterminetheprincipalaxisofanisolatedperson,andacquirethecorrespondingpositionofeachtargetindifferentcameras.Basedonsingle-cameratracking,ourmethodfirstacquirethehumanbody’sbinaryimagebybackgroundsubtraction,andthengettheaxisinformationofthetargetbytheLeastMedianofSquares,finallywerealizethedectectionofthetargetundermulti-cameraenvironmentbycalculatingthehomographybetweencamerasandthetarget’scorrespondingpositionindifferentcamerasacquiredbythehomography.Secondly,studiedamulti-cameramulti-targettrackingalgorithmbasedonMCMCparticlefilter.Thisalgorithmfirstestablishedautoregressivemotionmodelandobservationmodelbasedonthetargetcolorhistogramandmotionhistogram,usingparticlefilterfortrackingineachsinglecamera,andthenfusethedatafromcamerasincentralprocessorandtrackingthetargetsovermulti-camerasbasedontheMCMCparticlefilter.ThismethodcontrastedthelaboratorydataandthepublicdatasetsofCARIARandPETS,andcomparedwithotherclassicalmethods.Theexperimentresultsshowthatthemethodhasmoreadvantagesontherecallandprecisionthantraditionalmethods.Finally,weestablishedasmallmulti-targettrackingsystemunderalaboratoryenviro