简介：Videoobjectsegmentationisimportantforvideosurveillance,objecttracking,videoobjectrecognitionandvideoediting.Anadaptivevideosegmentationalgorithmbasedonhiddenconditionalrandomfields(HCRFs)isproposed,whichmodelsspatio-temporalconstraintsofvideosequence.Inordertoimprovethesegmentationquality,theweightsofspatio-temporalcon-straintsareadaptivelyupdatedbyon-linelearningforHCRFs.Shadowsarethefactorsaffectingsegmentationquality.Toseparateforegroundobjectsfromtheshadowstheycast,lineartransformforGaussiandistributionofthebackgroundisadoptedtomodeltheshadow.Theexperimentalresultsdemonstratedthattheerrorratioofouralgorithmisreducedby23%and19%respectively,comparedwiththeGaussianmixturemodel(GMM)andspatio-temporalMarkovrandomfields(MRFs).