Head tracking Using Multi-observation Models in particle Filters

摘要

A novel particle filter is presented in this paper which incorporates multi-observation models(MOMPF)to overcomemismatching of the tracked object feature and particle reference model．In brief,it unifies the tracked object different features withthe particle reference model，and thus it forms a group of particles with different models to track the object．Based on the changingdisplays in the feature CUeS ofthe object，particles with different models convert alternately in the tracking process．With this method,we call effectively avoid losing object during the whole tracking process．The paper theoretically proves that it is a novel methodwhich extends from the classical one．The advantage of the proposed method over the standard PF in the context．of head tracking isillustrated through computer simulafions．．