资源描述:
《光流的经典文献,运动跟踪专业的同学必读 Performance o.pdf》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、1PerformanceofOpticalFlowTechniques121J.L.Barron,D.J.FleetandS.S.Beauchemin12Dept.ofComputerScienceDept.ofComputingScienceUniversityofWesternOntarioQueen'sUniversityLondon,Ontario,N6A5B7Kingston,Ontario,K7L3N6AbstractWhiledierentoptical
owtechniquescontinuetoappear,thereha
2、sbeenalackofquantitativeevaluationofexistingmethods.Foracommonsetofrealandsyntheticimagesequences,wereporttheresultsofanumberofregularlycitedopti-cal
owtechniques,includinginstancesofdierential,matching,energy-basedandphase-basedmethods.Ourcomparisonsareprimarilyempirical,
3、andconcentrateontheaccuracy,reliabilityanddensityofthevelocitymeasurementstheyshowthatperformancecandiersignicantlyamongthetechniquesweimplemented.1IntroductionWithoutdoubt,afundamentalproblemintheprocessingofimagesequencesisthemea-surementofoptical
ow(orimagevelocity).T
4、hegoalistocomputeanapproximationtothe2-dmotioneld{aprojectionofthe3-dvelocitiesofsurfacepointsontotheimag-ingsurface{fromspatiotemporalpatternsofimageintensity[31,58].Oncecomputed,themeasurementsofimagevelocitycanbeusedforawidevarietyoftasksrangingfrompassivesceneinterpret
5、ationtoautonomous,activeexploration.Ofthese,taskssuchastheinferenceofegomotionandsurfacestructurerequirethatvelocitymeasurementsbeaccurateanddense,providingacloseapproximationtothe2-dmotioneld.Currenttechniquesrequirethatrelativeerrorsintheoptical
owbelessthan10%[10,36].Ve
6、rriandPoggio[58]havesuggestedthataccurateestimatesofthe2-dmotioneldaregen-erallyinaccessibleduetoinherentdierencesbetweenthe2-dmotioneldandintensityvariations,whileothers(e.g.[4])arguethatthemeasurementofoptical
owisanill-posedproblem.Forthesereasonsithasbeensuggestedtha
7、tonlyqualitativeinformationcanbeextracted.Manymethodsforcomputingoptical
owhavebeenproposed{otherscontinuetoappear.Lacking,however,arequantitativeevaluationsofexistingmethodsanddirectBarron,FleetandBeaucheminIJCV12:1,pp43-77,19942comparisonsonasinglesetofinputs.Kearneyetal.
8、[37]discussedsourcesoferrorwithgradient-basedmethods.LittleandVerri[39]comparedpro