Analysis of Broadcast Soccer Video, Trajectory-Based Ball Detection,and Tracking

The soccer imaging and broadcasting companies try to improve the audience’s satisfaction bysubstituting manual analysis methods of soccer videos with automatic ones, to provide more statistical realtimeinformation about the current match. One of the most important issues in automatic analysis of soccervideos is the detection of the ball and its trajectory, which is a more challenging task compared to other objecttracking problems. The difficulties emanate from the small size of the ball in long shots, ball deformation overframes, presence of more than one ball on the field or billboards, as well as merging the ball with the lines ofthe field, and occluding with players. In this paper, we propose a framework for detecting a ball and itstrajectory. In the proposed method, after playfield detection and shot filtering, ball candidates are detected inthe initial frames using morphology operators and object filtering. Then, candidate trajectories are extracted,scored, and ranked; and the trajectory with the best rank is selected. Later, the Kalman filter and blockmatching methods are used for detecting the ball and predicting its next positions in the next remaining frames.The proposed algorithm has been evaluated on ۱۱۲۵ video frames of the match between Esteghlal andPersepolis played at Azadi Stadium of the Islamic Republic of Iran in ۲۰۱۷. The results of experiments show۹۵.۸%, ۹۸.۸%, and ۹۶.۹% for the accuracy, precision, and recall criteria respectively

Keywords:

soccer video , Kalman filtering , Homography transformation , block matching , image processing , videoprocessing , ball tracking