A Stereoscopic Approach for the Association of People Tracks in Video Surveillance Systems

Abstract

analyses (COLLINS et al. 2001). The more accurate the object coordinates of people in the scene are known, the more detailed analyses can be conducted with respect to motion patterns or interactions between tracked people. In this paper we present an approach that generates consistent global tracks of people in non-crowded scenarios. The trajectories are calculated in a common reference frame from observations of multiple surveillance cameras. An important step in our work is the estimation of body height as well as a reliable association of tracks across partly overlapping views. For that purpose, stereoscopic analysis is applied to overlapping parts of images which are either generated randomly while scanning wide areas with several PTZ cam