Analysis-by-synthesis: Pedestrian tracking with crowd simulation models in a multi-camera video network

Integrate crowd simulators to a multi-camera tracking system and improves the performance.Compared two simulators. The one with a more realistic simulation strategy has better results.The experiments are conducted on a very challenging dataset for crowds with multiple views. For tracking systems consisting of multiple cameras with overlapping field-of-views, homography-based approaches are widely adopted to significantly reduce occlusions among pedestrians by sharing information among multiple views. However, in these approaches, the usage of information under real-world coordinates is only at a preliminary level. Therefore, in this paper, a multi-camera tracking system with integrated crowd simulation is proposed in order to explore the possibility to make homography information more helpful. Two crowd simulators with different simulation strategies are used to investigate the influence of the simulation strategy on the final tracking performance. The performance is evaluated by multiple object tracking precision and accuracy (MOTP and MOTA) metrics, for all the camera views and the results obtained under real-world coordinates. The experimental results demonstrate that crowd simulators boost the tracking performance significantly, especially for crowded scenes with higher density. In addition, a more realistic simulation strategy helps to further improve the overall tracking result.