资源描述：
《Multi-object detection and tracking by stereo vision.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、Multi-ObjectDetectionandTrackingbyStereoVisionLingCaia,LeiHe∗,b,YirenXua,YumingZhaoa,XinYangaatheDepartmentofAutomation,ShanghaiJiaoTongUniversity,Shanghai,200340China.btheNationalLibraryofMedicine,NationalInstitutesofHealth,Bethesda,MD20894,USA.AbstractThispaperp
2、resentsanewstereovision-basedmodelformulti-objectdetec-tionandtrackinginsurveillancesystems.Unlikemostexistingmonocularcamera-basedsystems,astereovisionsystemisconstructedinourmodeltoovercometheproblemsofilluminationvariation,shadowinterference,andobjectocclusion.
3、Ineachframe,asparsesetoffeaturepointsareidentifiedinthecameracoordinatesystem,andthenprojectedtothe2Dgroundplane.Akernel-basedclusteringalgorithmisproposedtogrouptheprojectedpointsaccordingtotheirheightvaluesandlocationsontheplane.Byproducingclusters,thenumber,posi
4、tion,andorientationofobjectsinthesurveillancescenecanbedeterminedforonlinemulti-objectdetectionandtracking.Ex-perimentsonbothindoorandoutdoorapplicationswithcomplexscenesshowtheadvantagesoftheproposedsystem.Keywords:Stereovision,kerneldensityestimation,multi-objec
5、tdetectionandtracking,clustering.CorrespondingauthorEmailaddresses:cailing.cs@gmail.com(LingCai),lei.he@nih.gov(LeiHe),xuyiren@gmail.com(YirenXu),arolazym@sjtu.edu.cn(YumingZhao),yangxin@sjtu.edu.cn(XinYang)PreprintsubmittedtoPatternRecognitionJune16,20101.Introd
6、uctionSurveillanceformultipleobjectsisanactiveresearchtopicincomputervision,whichiswidelyappliedinpublicsafety,trafficcontrol,andintelligenthuman-machineinteraction,tojustnameafewexamples.Asurveillancesystemusuallyconsistsoftwocorrelatedcomponents:objectdetectionand
7、tracking,bothofwhichhavebeenextensivelystudiedandnumerousap-proacheshavebeenproposed.Thedetectioncomponentistolocatetheobjectsofinterest,andthetrackingcomponentassociatesobjects’positionsovertimeinasequenceofframes.Thesetwocomponentsarecorrelatedwitheachother,i.e.
8、,objectdetectionlocatesregionsofinterestsfortracking,andtrackingresultscanbeusedforefficientdetectioninsubsequentframes.Cur-rentsurveillancesystemsusually
