Round Segmentation in Table Tennis Videos Using Player Tracking

• 1、School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, 100088

Abstract：This paper addresses the problem of round segmentation in table tennis match videos. Round segmentation aims to divide a match video sequence into multiple sub-clips based on round boundaries, providing a foundation for subsequent technical and tactical analysis and highlight extraction. We propose a round segmentation method that detects and tracks player targets in video frames, analyzing their position changes to determine segmentation moments between rounds. First, player targets are extracted and discriminative feature representations are generated using object detection and ReID techniques. Then, player targets are matched and tracked across frames to analyze position changes. Finally, segmentation moments and the number of rounds are determined based on the position changes. Experimental results demonstrate that the proposed method accurately detects round segmentation moments and achieves automatic round segmentation. This research offers new insights and methods for intelligent processing of table tennis match videos, with both theoretical and practical application value.

Keywords： artificial intelligence table tennis video analysis player tracking object detection