An efficient multi-cue fusion based sequential monte carlo method for image sequence-based tracking

Abstract

This paper presents an efficient image sequence tracking method based on multiple cues fusion in the sequential monte carlo method. we combine background-weighted color histogram with edge histogram into sequential monte carlo algorithm for tracking. The color-based histogram is robust against noise and partial occlusion, but suffers from the presence of the confusing colors in the background. So, background-weighted color histogram is used to describe objects color feature. The edge feature may provide complementary information for tracking as well. Color histograms and edge histograms are used to model the object observations likelihoods function. These observations are used to obtain a posterior probability distribution for the location of the object in the sequence images based on sequential Monte Carlo method. The experiments on real image sequences have shown that the combination of color for tracking with other image feature can achieve more robust tracking.