This study proposes an intelligent evaluation system for traditional sports movements based on Generative Artificial Intelligence (GAI) and Large-Scale Generative Adversarial Networks (BigGAN). The system constructs a multimodal analysis framework integrating biological features and kinematic features and establishes a systematic movement evaluation index system. Based on BigGAN, the system implements a movement generation and visual reconstruction module, to improve the intuitiveness and interpretability of evaluation. Experimental results show that, in the movements generated and evaluation results obtained by this system, the average absolute error of joint angles is 2.73°, the root mean square error of trajectories is 14.36 mm, the movement smoothness is 0.87, and the deviation of muscle synergy activation index is 0.082. These indicators are all superior to those of existing mainstream methods, including Dynamic Time Warping (DTW), Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN) and Generative Adversarial Network (GAN). The system has obtained recognition from professional coaches in terms of visual authenticity and error recognition accuracy. Its comprehensive scores for all movements are higher than 3.7. In addition, the system maintains stable performance under noise interference and its response time is controlled within 182 milliseconds. Compared with existing methods, this method achieves better performance in quantitative indicators, and further improves the systematicness, interpretability and practicability of movement evaluation through cross-modal fusion and high-fidelity visual feedback strategies. This study provides a new technical path for the intellectualization of sports training.
Zhang et al. (Wed,) studied this question.