Multiple Body Part Tracking Using a Probabilistic Data Association Filter

This paper presents a framework for multiple body part tracking based on a probabilistic data association (DA) filter. The body parts are extracted using iterative cluster background subtraction and foreground modeling with pictorial structures. The background subtracted silhouette is cluttered and the body parts are subject to occlusions. The main novelty of the paper is in the effective solution for data association which involves tracking body parts based on the expected likelihood method. We also show the advantage of the expected likelihood DA over the standard Probabilistic Data Association Filter (PDAF). A number of experiments have been conducted on several synthetic and real-time data sets and encouraging results have been obtained.