A Multi-target Tracking Algorithm Based on Multiple Cameras
基于多相机的多目标跟踪算法

The reliable tracking of multi-targets is a challenging issue in computer vision. In this paper, we propose a novel multi-target localizing and tracking algorithm based on multiple cameras. Firstly, the view-to-view homographies are computed using several landmarks on different planes. Then, the foreground likelihood map in each view is obtained by using a codebook background modeling algorithm. Finally, we can localize multiple objects at multiple planes and perform tracking by shortest paths optimization. Compared with other popular methods, our proposed algorithm does not require computing the vanishing points of cameras. Therefore, it reduces the complexity and improves the accuracy simultaneously. Adopting the shortest path optimization algorithm can improve the tracking efficiency. The experimental results demonstrate that our method is robust to occlusion and also can achieve real-time performance.