Intelligent evaluation algorithm for Taekwondo poomsae quality and posture recognition teaching system

• Innovative human key point detection model for Taekwondo education. • Achieved high accuracy in behavioral recognition algorithms. • Efficiently combines AI technology with traditional Taekwondo training. • Offers improved learning quality and coaching independence for students. Taekwondo is a traditional martial art that originated in South Korea, which not only promotes a healthy lifestyle, but also enhances self-discipline and confidence. The quality of Taekwondo education, especially poomsae (forms) practice, largely depends on subjective judgments by coaches or referees, resulting in inconsistent evaluations and lengthy processes. Therefore, the study proposes a dynamic human key point detection model based on Graph Convolutional Networks (GCN). This model simplifies human actions by constructing a spatiotemporal map of human key points, extracting behavioral features through the GCN, and identifying Taekwondo poomsae actions. Experimental results show that when the dataset size was 800, the accuracy of the designed model reached 0.98, which was better than that of other models such as the predictive encoding GCN (0.90), behavior structure GCN (0.88), and skeleton behavior recognition GCN (0.83). Furthermore, the designed model achieved a low Root Mean Square Error (RMSE) of 0.10, while other models had RMSE of 0.15, 0.19, and 0.28, respectively. The operation time of the designed model was 4.6 seconds, demonstrating its superior efficiency and accuracy in detecting and recognizing Taekwondo poomsae. This study demonstrates the feasibility of integrating artificial intelligence with Taekwondo teaching to enhance the quality of training.