文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > ROBUST OBSERVATIONS FOR OBJECT TRACKING
侵权投诉

ROBUST OBSERVATIONS FOR OBJECT TRACKING

It is a difficult task to find an observation model that will perform well for long-term visual tracking. In this paper, we propose an adaptive observation enhancement technique based on likelihood images which are derived from multiple visual features. Th

ROBUSTOBSERVATIONSFOROBJECTTRACKING

BohyungHanandLarryDavisDept.ofComputerScienceUniversityofMarylandCollegePark,MD20742,USA{bhhan,lsd}@cs.umd.edu

ABSTRACT

Itisadif culttaskto ndanobservationmodelthatwillperformwellforlong-termvisualtracking.Inthispaper,weproposeanadaptiveobservationenhancementtechniquebasedonlikelihoodimageswhicharederivedfrommultiplevisualfeatures.ThemostdiscriminativelikelihoodimageisextractedbyPrincipalComponentAnalysis(PCA)andincrementallyupdatedframebyframetoreducetemporaltrackingerror.Intheparticle lterframework,thefeasibil-ityofeachsampleiscomputedusingthismostdiscrimina-tivelikelihoodimagebeforetheobservationprocess.Inte-gralimageisemployedforef cientcomputationofthefea-sibilityofeachsample.Weillustratehowourenhancementtechniquecontributestomorerobustobservationsthroughdemonstrations.

1.INTRODUCTION

Trackersarebasedonthesomemeasurementofsimilaritybetweenthetargettobetrackedandobservations,andvar-iousobservationmethodsareusedtode nethissimilarity.Intensityorcolorisnaturaltouseinobjecttracking,andap-proachesbasedontemplates[1]andhistograms[2,3,4,5]areverycommon.Also,edges[8]and lterresponses[9,6]areimportantfeaturesforobjecttracking.Variousobserva-tionstrategieshavebeenproposed,butthereisnogenerallysuperiorobservationmethodforgeneralvisualtrackingal-gorithms.So,weinsteadproposeanobservationenhance-menttechniquebasedonlikelihoodimages,whichcanbeincorporatedintomanytrackingalgorithms.

Alikelihoodimagerepresentsthecontrastinformationbetweenforeground(targetregion)andbackground(itssur-rounding);itiscreatedbycomparinghistogramsofbothareaswithrespecttosomefeatures.Itwasoriginallysug-gestedin[10]fortrackingproblems;there,themostdis-criminativefeatureselectedfromasetoflikelihoodimagesisdirectlyusedformean-shifttracking.However,thisap-proachmayexhibitpoorperformanceinclutterandcanlosethetargetinspiteofitsvisualsalience.Also,thelikeli-hoodimagecanbesigni cantlycontaminatedbytemporary

trackingerrors.Therehavebeensomecloselyrelatedworks[7,6],buttheydonotprovideanadequatesolutiontotheseproblems.

Inthispaper,wepresentanobservationenhancementtechniqueusinglikelihoodimagesobtainedfromtwodif-ferentfeaturespaces–RGBandnormalizedRGB(rgb).Sixlikelihoodimagesarecreated,andamostdiscrimina-tivelikelihoodimageisextractedbyPCA.Inordertoavoidpollutionoftheextractedlikelihoodimagebytrackinger-ror,weupdatethesubspaceincrementallyontheassump-tionthatthescenearoundthetargetchangessmoothly.The“brightness”inthemostdiscriminativelikelihoodimagede-liverspriorinformationaboutthetargetregion,anditisem-ployedtocomputethefeasibilityofasample(tobede nedrigorouslybelow)beforethemeasurementstepinthepar-ticle lter.Thefeasibilityisobtainedbythesummationofvaluesinsidetheregioninthemostdiscriminativelikeli-hoodimage,andanintegralimage[12]isemployedforef- cientcomputation.The nalweightofeachparticleisde-terminedbytheproductofthefeasibilityandthelikelihoodinobservation.Asaresult,severalindependentfeaturesaremergedthroughthelikelihoodimagesandthemergedfea-turesareutilizedforrobustobservationviathefeasibilitycomputation.

Thispaperisorganizedasfollows.Wedescribethelike-lihoodimagesandthefeatureextractioninsection2and3,respectively.Insection4,trackinginaparticle lterframe-workandexperimentalresultsaredemonstrated.

2.LIKELIHOODIMAGES

Likelihoodimagesrepresentthedistinctivenessofatargetobjectfrombackgroundwithrespecttoagivenfeatureorsetoffeatures.Fortheconstructionoflikelihoodimages,log-likelihoodratiosareobtained rstfromhistogramsofforegroundandbackgroundpixels.Then,thesalientregioninforegroundcanbedetectedbyidentifyinghighlikelihoodratios.

Indetail,supposetheforegroundisgivenandtheback-groundisregardedasarectangularregionsurroundingthe

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email