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 > 工程科技 > 电子/电路 > Stable Multi-Target Tracking in Real-Time Surveillance Video
侵权投诉

Stable Multi-Target Tracking in Real-Time Surveillance Video

StableMulti-TargetTrackinginReal-TimeSurveillanceVideo

BenBenfoldandIanReidDepartmentofEngineeringScience

UniversityofOxford

Oxford,UK

{bbenfold,ian}@robots.ox.ac.uk

Abstract

Themajorityofexistingpedestriantrackersconcentrateonmaintainingtheidentitiesoftargets,howeversystemsforremotebiometricanalysisoractivityrecognitioninsurveil-lancevideooftenrequirestablebounding-boxesaroundpedestriansratherthanapproximatelocations.Wepresentamulti-targettrackingsystemthatisdesignedspeci callyfortheprovisionofstableandaccurateheadlocationes-timates.Byperformingdataassociationoveraslidingwindowofframes,weareabletocorrectmanydataas-sociationerrorsand llingapswhereobservationsaremissed.Theapproachismulti-threadedandcombinesasyn-chronousHOGdetectionswithsimultaneousKLTtrackingandMarkov-ChainMonte-CarloDataAssociation(MCM-CDA)toprovideguaranteedreal-timetrackinginhighdef-initionvideo.Wherepreviousapproacheshaveusedad-hocmodelsfordataassociation,weuseamoreprincipledapproachbasedonaMinimalDescriptionLength(MDL)objectivewhichaccuratelymodelstheaf nitybetweenob-servations.Wedemonstratebyqualitativeandquantita-tiveevaluationthatthesystemiscapableofprovidingpre-ciselocationestimatesforlargecrowdsofpedestriansinreal-time.Tofacilitatefutureperformancecomparisons,wemakeanewdatasetwithhandannotatedgroundtruthheadlocationspubliclyavailable.

Figure1.Anexampleofaframeinwhichwewouldliketotrack(top)andsampleoutputfromoursystem(bottomleft).Theim-agesinthebottomrightshowtheresultofanaiveapproachwhichappliesa xedoffsettoapedestriandetectiontoestimatetheheadlocation,theresultofwhichisbadlycentredanddriftsaroundasthepedestrianwalks.

1.Introduction

Theperformanceofpedestriantrackingsystemshassteadilyincreasedtothepointwhereattemptsarebeingmadetotrackdensecrowdsofhighlyoccludedpedestri-ans.Theavailabilityofhighresolutionsurveillancevideohasopenedupnewopportunitiesforexploitingtheoutputofsuchtrackersbyusingthemtoproducestreamsofhighresolutionpedestrianimages.Theseimagescanbefurtheranalysedtoestimatebodypose,recogniseactionsorforpassivefacerecognition.

Inthispaperwedescribeasystemwhichisdesigned

Stable Multi-Target Tracking in Real-Time Surveillance Video

ex-plicitlyfortheprovisionofaccuratepedestrianheadloca-tionsthataresuitableforfurtherprocessing.Ourgoalistoachieverobustnessandhighlevelsofaccuracywhilstmain-tainingreal-time(25fps)performancewhentrackingmul-tiplepedestriansinhighde nitionsurveillancevideo.Theadvantagesoftrackingheadsdirectlytoobtainstableim-agesareillustratedin gure1.Inthecontextofgeneralpedestriantracking,usingtheheadasthepointofrefer-enceratherthantheentirebodycanbemorereliablebe-causeheadsarerarelyobscuredfromoverheadsurveillancecamerasandaregenerallynotobscuredbyclothing.Mostrecentworkonmultipletargettrackinghasfocused

Stable Multi-Target Tracking in Real-Time Surveillance Video

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