Segmentingindividualhumansinahigh-densityscene(e.g.,acrowd)acquiredfromastaticcameraischallengingmainly duetoobjectinter-occlusion(Fig.1).We de
ne this proble mas
a“model-asedsegmentation”problemandthesolutionisob-
tainedusingaMarkovchainMonteCarlo(MCMC)approach.
Knowledgeofvariousaspectsincludinghumanshape,human
height,cameramodel,andimagecuesincludinghumanhead
candidates,foreground/backgroundseparationareintegrated
inaBayesianframework.Weshowpromisingresultsonsome
challengingdata.