AI vision-based athlete pose analysis for training assistance in badminton

This project is expected to develop an assistant system suitable for basic training in badminton footwork and batting actions in two years. The main algorithm of our project is a human pose/posture estimation and body keypoints positioning technique based on deep learning. According to the technical context and the cooperation between subjects, we divide our project into two main phases as follows. The first phase is to research the main axis of the project, such as human pose estimation, basic footwork analysis, and batting action recognition, during the first year. The works include: 1) building a stereo vision system for half-court monitoring and completing the camera calibration; 2) researching human pose estimation and body keypoints positioning. The innovative difference from existing methods is that racket detection is added aiming at badminton sports and it is attempted to be integrated into the collection of body keypoints of human pose. Additionally, the normal vector of the racket face is derived to extend advanced research of the second year. 3) completing basis and some advanced footwork recognition: visualizing movement trajectory and forming a heatmap based on the centroid of the player. 4) completing basic batting action recognition: a total of 7 actions will be considered. The action difference between players and coaches is calculated to facilitate adjustment to a stable state. The second phase is the extended application research, including: 1) completing the research on the full-court monitoring system and video synchronization; 2) completing the analysis of advanced footwork and batting actions. In this part, the shuttlecock tracking by subproject 2 and advantageous region by subproject 4 are combined with our research results. 3) conducting onsite verification and system integration: the goal of the master project will be achieved under the coordination of the host of the master project. During the final stage, badminton players are invited to participate in system evaluation and improvement. Consequently, we hope that the developed system will be practical.