Tai Chi movement quality evaluation model based on adaptive multi scale ST-GCN

With the increase of health and fitness awareness, Tai Chi, a traditional sport, has gradually attracted attention, and the assessment of its movement quality has become an important area of academic research. However, the complexity of Tai Chi movements and the variety of rhythmic changes make automated assessment of them a significant challenge. Therefore, the study first improved the OpenPose bone extraction method. Second, an action feature extraction model based on an adaptive multi-scale spatial-temporal graph convolutional network (ST-GCN) was designed. The model was designed to cope with Tai Chi movement features with different movement amplitudes and rhythms by introducing an adaptive multi-scale mechanism. It also combined with the spatial relationship modeling capability of the graph convolutional network to effectively capture the motion information of the key parts of the human body. In the improved OpenPose test, its mean average precision values on the two datasets were 82.4% and 85.1%, respectively. The percentage of correct keypoints for the two complex joint parts of the knee and ankle were 83.5% and 82.0%, respectively, and the model complexity was only 12.6. The Top-1 and Top-5 accuracy of the improved motion feature extraction model were improved by 7.1% and 4.2%, respectively. When the number of samples was 5000, the selection feature extraction accuracy and mean absolute error were 94.7% and 5.2 pixels, respectively. The correlation coefficient of quality score under different lighting conditions was 0.91. In contrast to the traditional model, the experimental results demonstrate that the model has excellent robustness and accuracy in the extraction of multi-scale action features and the discrimination of action accuracy, improving the accuracy of action quality assessment. At the same time, the model has good adaptability and stability in complex scenarios and can be applied to a variety of practical application scenarios. The proposed ST-GCN model can provide a new technical means and theoretical support for the intelligent assessment of traditional sports.