People tracking with multi-camera system

This paper presents a method for tracking people using multiple cameras. The system is implemented with a two level processing strategy. In low-level, object trajectories are detected on each camera image sequence (track detection). This procedure involves active region extraction and matching. In high-level, all the trajectories extracted from the multi camera system are related in order to create a global view (track matching). This is accomplished by homography transformations between image planes. The total set of detected trajectories and there relations is represented by a graph. Experimental results are preformed with recorded data sets and PETS2001 sequence.