Tracking Groups of People in Presence of Occlusion

This paper addresses the problem of people group tracking in presence of occlusion as people form groups, interact within groups or leave groups. Foreground objects (a person or a group of people) from two consecutive frames are matched based on appearance (RGB histogram) and object location (2D region) similarity. While tracking, this method determines and handles some events such as objects merging and splitting using forward and backward matching matrices. The experimental results show that the proposed algorithm is efficient to track group of people in cluttered and complex environments even when total or partial occlusion occurs.