基于低秩-稀疏联合表示的视频序列运动目标检测（英文）.docx

基于低秩-稀疏联合表示的视频序列运动目标检测（英文）
Abstract
Motionobjectdetectionisacriticaltaskinvideoanalysisanddetectionfield.Theexistingmotiondetectionmethodshavelimitationsindealingwithcomplexscenes.Therefore,thispaperproposesamotionobjectdetectionmethodbasedonlow-rankandsparsejointrepresentationforvideosequence,whichisabletocapturethespatiotemporalcharacteristicsofthemotionobjectinamoreaccurateandrobustway.Theproposedmethodfirstextractsthemotionfeaturefromthevideosequenceandthenproposesajointrepresentationmodelusinglow-rankandsparseapproximation.Experimentalresultsonpubliclyavailabledatasetsdemonstratethattheproposedmethodoutperformsthestate-of-the-artmethodsandachieveshighdetectionaccuracywithlowcomputationalcost.
1.Introduction
Motionobjectdetectionplaysasignificantroleinmanycomputervisionapplications,suchassurveillance,trafficanalysis,andhuman-computerinteraction.Withthewideavailabilityofvideodata,automaticmotionobjectdetectionhasbecomeanattractiveandimportantresearcharea.Mostexistingmotiondetectionmethodsuseacombinationofmotionfeatureextractionandclassificationtechniques.However,thecomplexityandvariabilityofmotionpatternsintherealworldposesignificantchallenges.Manyexistingmethodscannotcapturethespatiotemporalcharacteristicsofmotionobjectsaccuratelyandrobustly.Oneofthereasonsforthisisthatthemotioninformationisoftennoisyandredundant.
Toaddresstheseissues,thispaperproposesamotionobjectdetectionmethodbasedonlow-rankandsparsejointrepresentationforvideosequence.Theproposedmethodismotivatedbyrecentadvancesinthetheoryofsparserepresentationandlow-rankapproximation.Theproposedmethodfirstextractsthemotionfeaturefromthevideosequenceandthenproposesajointrepresentationmodelusinglow-rankandsparseapproximation.Thismodelisabletocapturethespatiotemporalvariationsofmotionobjectsmoreaccuratelyandrobustly.Toevaluatetheperformanceoftheproposedmethod,aseriesofexperimentswereconductedonpubliclyavailabledatasets.Theresultsdemonstratethattheproposedmethodoutperformsthestate-of-the-artmethodsandachieveshig