Event detection for broadcast tennis videos based on trajectory analysis

This paper presents an effective and efficient semantic event detection method for broadcast tennis videos. We proposed an effective player tracking and trajectory modification technique, and then event detection based on trajectory analysis is developed accordingly. In our work, the playfield and court line are extracted from a court view. Then, we detect player objects and utilize adaptive Kalman filter to improve the tracking accuracy. In event detection phase, we refine the unreasonable trajectory to avoid the noise interference especially in upper court, and eliminate the minor clip which can´t be a meaningful event. Finally, we analyze the players´ trajectory combined with tennis rules to extract semantic events. The experiments indicate that the proposed method achieves satisfactory performance with 80% of precision rate and 89% of recall rate.