Multi-Camera People Detection and Tracking

Abstract

Pedestrian detection and tracking are essential problems in the field of computer vision, having a wide range of applications in surveillance, security, autonomous driving and robotics areas. Although people detection and tracking are generally considered widely-used technologies currently, however, occlusions still remain a major challenge. In this paper, we propose an approach to improve the detection and tracking performance in multi-camera scenarios with overlapping field-of-views, which allows for better handling of occlusion problem. It mainly includes monocular people detection, projection, fusion, probabilistic occupancy map generation and multi-object tracking steps. Evaluations for detection and tracking on WILDTRACK dataset have been implemented and the results indicate that our approach outperforms state-of-the-art deep learning methods which haven’t been trained on WILDTRACK.