Skip to Main Content
The tracking robustness can be improved greatly using multi-camera system because the overlapping field of view provides more information of the objects. However, traditional template matching based multi-camera object tracking suffers from the tracking errors brought by view point differences. To solve the problem, we propose the extended online HECOL (Homography and Epipolar-based Consistent Labeling) to communicate templates between different camera views which is no longer limited to the ground plane. We use the Structural Similarity Image Measurement (SSIM) as the criteria to match between different views. Besides, we applied the Gaussian model and Bayesian statistics to fuse different cameras' tracking results of last moment. Experiments show that the proposed way of transferring different view points' template reduces tracking error when occlusion and 3D rotation occur.