Multi-target Tracking via Max-Entropy Target Selection and Heterogeneous Camera Fusion

Nowadays, dual-camera systems, which consist of a static camera and a pan-tilt-zoom (PTZ) camera, have become popular in video surveillance, since they can offer wide area coverage and highly detailed images of the interesting starget simultaneously. Different from most previous multi-target tracking methods without information fusion, we propose a multi-target tracking framework based on information fusion of the heterogeneous cameras. Specifically, a conservative online multi-target tracking method is introduced to generate reliable tracklets in both cameras in real time. A max-entropy target selection strategy is proposed to determine which target should be observed by the PTZ camera at a higher resolution to reduce the ambiguity of multi-target tracking. Finally, the information from the static camera and the PTZ camera is fused into a tracking-by-detection framework for more robust multi-target tracking. The proposed method is tested in an outdoor scene, and the experimental results show that our method significantly improves the multi-target tracking performance.