Tracking deformable structures, with no prior on their possible shapes, is a very chal- lenging problem. Indeed, the shape of a deformable object may change drastically between two consecutive images within a video sequence. These deformations are due to object apparent motion, to perspective effects and to 3D shape evolution. This difficulty is amplified when the object becomes partially or totally occluded during even a very short time period. The presence of cluttered background and ambiguities constitutes other difficulties for tracking. This problem has been treated quite extensively in the computer vision literature, and there are several different algorithms which have been developed to address it. These schemes are mainly based either on the snakes model or level-set approaches. In this report, we propose an alternative approach based on a parametric level set function. To perform the sequential inference on this difficult problem, we propose a Markov Chain Monte Carlo (MCMC)-based Particle algorithm.
