Badminton Video Analysis Based on Player Tracking and Pose Trajectory Estimation

The traditional attitude estimation method is still not ideal for occlusion, which often ignores the continuity of human posture itself and the spatiotemporal continuity of human target trajectory. Therefore, this paper considers using the improved CAMSHIFT algorithm to track the athletes, taking badminton as an example, proposes a human posture estimation algorithm based on video. The scheme uses H, S, V to establish the athletes' color model. The traditional histogram is replaced by weighted histogram and instance histogram. The algorithm also estimates the pose of the human body in static image, and establishes the human body tracking based on the measurement of the pose distance between frames. It generates better candidate keys for the key points with lower confidence in each frame, and generates the optimal global pose for each frame by recombining these body parts. Finally, the real-time semantic analysis scheme of badminton video is formed by using the above methods, and the performance indexes of the whole system are comprehensively analyzed through experiments. It ensures the tracking processing speed and reduces the error caused by similar tracking targets, which proves the feasibility and effectiveness