Robust body parts tracking using particle filter and dynamic template

This paper presents a novel body parts tracking algorithm using particle filters and a dynamic head template to handle occlusions. Hands and head tracking is a challenging problem because of the similarity of colour, the presence of noise and occlusions. Two different particle models are considered; a pixel and a rectangular region. Head is modelled as a rectangle and hands as a cloud of points because of the high hand shape variability. The exclusion principle is used to allow particle filters to interact between them to handle occlusions. Each filter sample set is used to penalize the observations of any other filter. For the head particle filter, we use a dynamic template to determine the penalizing coefficients. Experiments have been performed and have shown high performances when compared with other trackers.