Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf

Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf

ID:48126451

大小:2.25 MB

页数:9页

时间:2019-11-27

Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf_第1页
Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf_第2页
Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf_第3页
Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf_第4页
Chari_On_Pairwise_Costs_2015_CVPR_paper.pdf_第5页
资源描述:

《Chari_On_Pairwise_Costs_2015_CVPR_paper.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、OnPairwiseCostsforNetworkFlowMulti-ObjectTrackingViseshChariSimonLacoste-JulienyIvanLaptevJosefSivicINRIAandEcoleNormaleSuperieure,Paris,France´AbstractMulti-objecttrackinghasbeenrecentlyapproachedwiththemin-costnetworkflowoptimizationtechniques.Suchmethodssimultaneouslyresolvemultipleobjectt

2、racksinavideoandenablemodelingofdependenciesamongtracks.Min-costnetworkflowmethodsalsofitwellwithinthe“tracking-by-detection”paradigmwhereobjecttrajecto-riesareobtainedbyconnectingper-frameoutputsofanob-(a)Nooverlapterm(b)Withoverlaptermjectdetector.Objectdetectors,however,oftenfailduetoocclusion

3、sandclutterinthevideo.Tocopewithsuchsit-uations,weproposetoaddpairwisecoststothemin-costnetworkflowframework.WhileintegersolutionstosuchaproblembecomeNP-hard,wedesignaconvexrelaxationsolutionwithanefficientroundingheuristicwhichempir-icallygivescertificatesofsmallsuboptimality.Weevalu-(c)Noco-occu

4、rrenceterm(d)Withco-occurrencetermatetwoparticulartypesofpairwisecostsanddemonstrateimprovementsoverrecenttrackingmethodsinreal-worldFigure1:Resultsofnetworkflowtrackingusingcostfunctionsvideosequences.with/withoutpairwiseterms.(a)-(b):apairwisetermthatpenal-izestheoverlapbetweendifferenttracksh

5、elpsresolvingambigu-1.Introductionoustracks(showninred)incrowdedscenes.(c)-(d):apairwisetermthatencouragestheconsistencybetweentwosignals(hereThetaskofvisualmulti-objecttrackingistorecoverheaddetectionsandbodydetections)helpseliminatingfailuresspatio-temporaltrajectoriesforanumberofobjectsina(s

6、howninred)ofobjectdetectors.videosequence.Trackingmultipleobjects,likepeopleorvehicles,hasawiderangeofapplicationsfromRoboticsproblem,min-costnetworkflow[2]isparticularlyattrac-tovideosurveillance[28].Despiterecentprogressinthetiveasitallowsforoptimalandefficientsolutions[22].field[3,5,8,20,21,22,

7、27],trackingremainsachallengingTheenergyminimizationapproachtotrackingenablesproblemespeciallyincrowdedandclutteredscenes.globalsolutionstotrackselectionandavoidsearlyandWiththeadvancesinobjectdetection,“tracking-by-error-proneloc

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。
相关文章
更多
相关标签