A Real Time Player Tracking System for Broadcast Tennis Video

This paper proposes a novel framework for tennis player detection and tracking. The algorithm is built on (1) a powerful court-line pixel detection method utilizing intensity and texture pattern, (2) a fast RANSAC-based line parameter estimation which also determines line extents, and (3) a player segmentation and tracking algorithm exploiting knowledge of tennis court model. The content of the video is then explored at a highly semantic level. The framework was tested extensively on numerous challenging video sequences with various court environments and lighting conditions. The results show the robustness and the promising direction of our algorithm.