预览加载中,请您耐心等待几秒...
1/2
2/2

在线预览结束,喜欢就下载吧,查找使用更方便

如果您无法下载资料,请参考说明:

1、部分资料下载需要金币,请确保您的账户上有足够的金币

2、已购买过的文档,再次下载不重复扣费

3、资料包下载后请先用软件解压,在使用对应软件打开

视频监控中目标检测与跟踪算法的研究的中期报告 [本文以英语回答,如有翻译不当,敬请谅解] Mid-termReportonTargetDetectionandTrackingAlgorithmsinVideoSurveillance Introduction: Thevideosurveillancesystemiswidelyusedinvariousscenarios,suchaspublicsecurity,transportation,andindustry.Targetdetectionandtrackingaretwoessentialcomponentsofthevideosurveillancesystem.Targetdetectionaimstolocateandrecognizetheobjectsofinterestinthevideoframes,whilethetargettrackingaimstotrackthetrajectoryofthedetectedobjectsacrossthevideoframes. Objectives: Themaingoalofthisresearchistoinvestigateanddevelopefficienttargetdetectionandtrackingalgorithmssuitableforvideosurveillance.Specifically,theobjectivesareasfollows: 1.Toreviewtherelatedliteratureontargetdetectionandtrackingalgorithms. 2.Toproposeanoveltargetdetectionalgorithmthathasbetteraccuracyandspeedthantheexistingalgorithms. 3.Todeveloparobustandeffectivetargettrackingalgorithmthatcanhandleocclusion,targetappearancechangesandotherchallenges. 4.Toevaluatetheproposedalgorithmsonbenchmarkdatasetsandreal-worldvideosurveillancescenarios. Progress: Wehavecompletedacomprehensiveliteraturereviewonthestate-of-the-arttargetdetectionandtrackingalgorithms,includingdeeplearning-basedmethods,multi-objecttracking,andvisualobjecttracking.Wefoundthatdeeplearning-basedmethodshaveachievedstate-of-the-artperformanceinobjectdetection,buttheprocessingspeedisstillachallenge.Wealsofoundthatmulti-objecttrackingandvisualobjecttrackinghavemadesignificantprogressinhandlingocclusionandtargetappearancechanges. WehaveproposedanoveltargetdetectionalgorithmthatcombinestheFasterR-CNNandYOLOv3architectures.Theproposedalgorithmachievesbetteraccuracythantheindividualmodelsandcanprocessthevideoframesatareal-timespeed.WealsodevelopedanonlinetrackingalgorithmthatintegratesthecorrelationfilterandsequentialMonteCarlomethods.Theproposedalgorithmcantracktheobjectsinrealtimeandhandleocclusionandtargetappearancechanges. NextSteps: Weplantoevaluatetheproposedalgorithmsonbenchmarkdatasetsandreal-worldvideosurveillancescenarios