In a shopping mall context, tracking is focused on humans, which usually have irregular and abrupt movements while they
are observing or trying on clothes or other products. For this reason, when an occlusion ends, the people can not be re-identified conveniently using predictions of their positions. In order to minimize the occlusion problems related to the non-regular trajectories of people in the studied environment, an occlusion management algorithm based on visual appearance is suggestedin this paper. This method complements the LSAP solution proposed for the detections-to-tracks association, and the combination of both gives as a result a powerful and robust tracking system oriented to video-surveillance of human behaviors.