Traditional table tennis technique evaluation relies heavily on coaches’ subjective judgment, which limits the objectivity, consistency, and scalability of instructional feedback. To address this problem, this study proposes a multi-dimensional joint quantitative evaluation method for table tennis techniques based on OpenPose and YOLOv8 using consumer-grade high-frame-rate video. A total of 50 participants were recruited and divided into a high-level group and a low-level group. Standardized forehand drive and backhand push tasks were recorded using a synchronized dual-view camera setup. OpenPose was used to extract upper-body keypoint trajectories for kinematic analysis, while YOLOv8 was employed to detect and track the ball, racket, and net for outcome-related feature extraction. Based on these data, seven core indicators covering movement stability, coordination, timing, smoothness, and hitting effectiveness were selected to construct a quantitative scoring model, which was further optimized by ridge regression and validated against expert ratings from three senior athletes/coaches. The results show significant between-group differences in multiple technical dimensions, including impact accuracy, smoothness, trajectory consistency, and limb coordination (p<0.001). The model score was strongly correlated with expert ratings (r=0.882, p<0.001) and demonstrated high reliability (ICC=0.915). These findings indicate that the proposed framework can provide a low-cost, non-invasive, and practically effective solution for intelligent table tennis teaching, technical diagnosis, and skill-level evaluation.
Yang et al. (Fri,) studied this question.