Efficient framework for extended visual object tracking

An algorithm for extending the Bayesian multiple target tracking framework to solve the extended visual object tracking problem using sparse features is proposed. In particular, the state space is divided into two sets: one modeling the global motion of the object and one modeling the movement of every feature point. This division allows one to obtain a factorized proposal distribution that, takes into account current measurements and exploits the structure of the problem, allowing an efficient exploration of the state space. The proposed method is demonstrated to be more accurate than the baseline algorithm while requiring lower processing time for the same performance.