Basketball dribbling action recognition and performance evaluation based on YOLO-Pose and LSTM

Basketball dribbling maneuvers are fundamental skills that require precise temporal-spatial coordination and technique assessment. This paper proposes an end-to-end framework for recognizing and evaluating dribbling actions using YOLO-Pose for real-time pose estimation and Long Short-Term Memory (LSTM) networks for temporal sequence modeling. We extract 17 keypoint coordinates from video frames and construct pose sequences that capture the kinematic characteristics of dribbling movements. The LSTM network learns the temporal dependencies between consecutive poses to classify action types and evaluate execution quality. Experimental results on a custom basketball dataset demonstrate that our method achieves 94.3% accuracy in action recognition and provides quantitative performance scores with high correlation to expert assessments. Compared to traditional methods, our approach offers superior real-time performance and robustness to background variations and player diversity.