Visual Tracking of Players through Occlusions in Low Resolution

F. Qi, Y. Luo, and D. Hu (PRC)


Image Processing and Applications, Object Tracking, Occlusion, Kalman Filter, Data Association


This paper presents a framework for tracking sports play ers in videos recorded by static domestic cameras. In the proposed approach, motion and appearance model is uti lized to tracking objects as details of the objects appear ance is hard to obtain in low resolution videos. To assign observed objects to a trajectory, a matching process base upon the motion and appearance model is provided. In the presence of occlusion, the data association may fail to as sign objects to some tracks. In this situation, the confliction is detected and some patches are generated to acquire more information to eliminate the puzzles. Then the trajectories are updated with or without observations. In the final step of tracking, new tracks are formed and some trajectories are deleted. As the foundation of tracking, Gaussian back ground model is utilized to segment out the players. The proposed method has been tested on real soccer game videos. The test results show that the system can successfully track multiple players independently and keep trajectories when occlusions occur.

Important Links:

Go Back