资源描述:
《Fragments Based Parametric Tracking》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、FragmentsBasedParametricTrackingPrakashC,BalamanoharPaluri,NalinPradeepS,andHiteshShahSarnoffInnovativeTechnologiesPrivateLimited,Ashaarch,MagrathRoad,Bangalore-560025,IndiaAbstract.Thepaperproposesaparametricapproachforcolorbasedtracking.Themethodfragmen
2、tsamultimodalcolorobjectintomultiplehomogeneous,unimodal,fragments.Thefragmentationprocessconsistsofmultilevelthresholdingoftheobjectcolorspacefollowedbyanas-sembling.Eachhomogeneousregionisthenmodelledusingasinglepara-metricdistributionandthetrackingisa
3、chievedbyfusingtheresultsofthemultipleparametricdistributions.Theadvantageofthemethodliesintrackingcomplexobjectswithpartialocclusionsandvariousdefor-mationslikenon-rigid,orientationandscalechanges.Weevaluatetheperformanceoftheproposedapproachonstandarda
4、ndchallengingrealworlddatasets.1IntroductionTwoprominentcomponentsofatrackingsystemare:objectdescriptorandsearchmechanism.Objectdescriptoristherepresentationoftheobjecttobetrackedusingasetoffeaturesthatcapturevariouspropertiesoftheobjectsuchastheappearan
5、ce,shape,textureetc.Givenanobjectdescriptor,thesearchmechanismlike[1,2],locatestheregioninanewimagethatbestmatchestheobjectdescription.Therearemultiplemethodssuggestedintheliteratureforobjectdescriptors.Mostofthesuccessfulmethodsfortrackingemploynon-para
6、metricobjectde-scriptorlikehistogram[1,3,4,5,6,7],asitfaithfullycapturesthevariabilityinthefeaturesoftheobjecttobetracked.However,withtheincreaseinnumberofobjectstobetrackedorthefeaturestobeconsidered,thehistogramsizegrowsexponentiallywhichisanundesiredb
7、ehavior.Toaddressthisissue,weproposeaparametricobjectdescriptorforcolorbasedtracking.AnN-dimensionalGaussiandistributionisemployedastheobjectdescriptorintheproposedapproach.Suchadescriptorcanaccuratelymodelaunimodalobject.Butobjectsunderconsiderationfort
8、rackingaregenerallymultimodalincolorspacemakingN-dGaussiandescriptorinsufficient.Hence,weneedtoconvertmultimodalobjectstounimodalrepresentation.Primarily,therearetwowaystoachievethisconversion:–Byprojectingthemultimodalobjec