AUTOMATIC TRACKING METHOD FOR SPORTS VIDEO ANALYSIS

摘要

This paper addresses the implementation of an advanced visual-based tennis sports video analysis system. Firstly, a white pixel-fraction based method is used, to detect the significant frames that include the tennis court. In addition, we employ the temporal correlation between two consecutive frames to track the court location within a local search area. Furthermore, we propose a player segmentation and tracking algorithm, that separately builds background models for the playing field and the area surrounding the field according to their different colors. This enables to provide it as a prediction of the current frame for more efficiency. Results on real tennis video data are presented, demonstrating the validity and performance of the approach.