An efficient prediction scheme for pedestrian tracking with cascade particle filter and its implementation on Cell/B.E.

Cascade Particle Filter was proposed for accurate object recognition in low frame rate video. However, Cascade Particle Filter can be expected to enhance the accuracy of recognition even in a regular frame rate video because of its run-time learning procedure. To apply such cascade particle filter for pedestrian recognition on surveillance and automotive applications, we propose an efficient prediction scheme optimized for pedestrian tracking in such applications. Moreover, we implement proposed scheme on Cell/B.E., one of the latest embedded high performance processors, to demonstrate real-time pedestrian tracking on embedded systems. Experimental result shows that proposed scheme improves pedestrian tracking accuracy by 22% with real-time processing on 30 fps video.