A Tennis Motion Correction Approach Based on Ensemble Learning and MediaPipe

This paper investigates human posture recognition methods in tennis sports and develops a tennis motion correction approach, which is capable of rectifying non-standard movements. Traditional tennis player posture detection methods suffer from several limitations, including insufficient robustness in complex backgrounds, high self-occlusion in tennis motions, and slow processing speeds for video-based action analysis. To address these issues, this paper proposes an integrated approach combining ensemble learning with the MediaPipe pose detection algorithm to address these challenges. This approach utilizes training data collected by an indoor motion capture system to train a tennis fundamental motion classification model based on Gradient Boosting Decision Trees (GBDT). MediaPipe is employed to perform human skeleton analysis, extracting eight key body joints. This paper evaluates tennis motions based on eight tennis-specific kinematic features and ultimately provides tailored corrective recommendations according to identified deficiencies. Experimental results demonstrate that this motion correction approach effectively delivers reasonable corrections for tennis players across different skill levels.