Multi-feature based high-speed ball shape target tracking

Ball shape object tracking is very important in sports video analysis such as tennis, golf ball and basketball. In this paper, a real-time high speed moving ball shape object tracking algorithm based on frame difference and multi-feature fusion is proposed. First of all, it detects the frame difference between the adjacent two frames. Then it divides the difference image into several small contours and decides if they are moving ball shape object areas by multi-feature based algorithm. It labelled the ball shape object and the smallest rectangle ROI (Region Of Interesting) which contains the moving object. At last, according to the location and size of ROI, intelligently control the size and location of CMOS image senor´s ROI. We also proposed a system to implement and validate the proposed algorithm in a wireless pan-tilt camera system. Experimental results show that the algorithm can be applied to real-time tracking of ball shape object with a high recognition rate, and the ROI control method can enhance the processing efficiency.