Adaptive tracking with patches and a new particle filter

This paper presents a new tracking method based on color histogram model and patch appearance model. We propose a new particle filter which processes tracking concurrently in several routes, and each route creates particles separately. Each particle is optimized with a new regular iteration method. Then we evaluate each particle in two levels: the whole level and the patch level. The two levels use color histogram and patch images to evaluate the particles respectively. Different routes have different appearance models. We also maintain a combined appearance model. Experimental results demonstrated the effectiveness of our method.