Occlusion Robust Tracking of Multiple Objects

This paper focuses on the problem of vision-based tracking of multiple objects. Probabilistic tracking in 3D supported by multiple video streams allows us to formalize an efficient observation model that is robust to occlusions. Each tracked object is assigned a support layer, a probabilistically meaningful pixel occupancy map, supplying weights used in the calculation of other objects observation likelihood. A Particle Filter implementation demonstrates the robustness of the resulting tracking system on synthetic data.