Multi-Task Tennis Stroke Biomechanics Analysis Using MediaPipe Pose
Researchers have developed a multi-task pipeline for analyzing tennis stroke biomechanics using only RGB video. The system automatically identifies strokes and predicts shot direction and posture quality, offering coaching tips based on a rule-based feedback layer. Utilizing MediaPipe Pose for landmark detection and a custom transformer model called TennisTransformerGPU, the pipeline achieves high accuracy in stroke-type recognition, even when trained on professional players and evaluated on amateurs. The study highlights the critical importance of using metric world coordinates for accurate cross-player analysis. AI