Simulation and optimization of Wushu Sanshou athletes’ movements based on generative adversarial networks

Wushu Sanshou is a competitive martial art that requires rapid, complex movements involving intricate coordination and biomechanical precision. Traditional methods for analyzing or simulating Sanshou techniques often fail to capture their full dynamism, making it difficult to enhance training outcomes or develop realistic motion systems. This research aims to simulate and optimize Wushu Sanshou athletes’ movements using a Temporal Convolutional Generative Adversarial Network with Modified Archimedes Optimization (TC-GAN-MA) method that captures realistic spatiotemporal motion patterns and refines them for biomechanical and performance effectiveness. High-frame-rate motion capture data from professional Wushu Sanshou athletes is collected and preprocessed. The raw video was converted into 3D joint sequences by extracting coordinates for all major joints. Kalman filtering was applied to reduce trajectory noise, followed by normalization to ensure consistency. The TC-GAN was employed to learn the temporal and spatial dynamics of Sanshou movements and synthesize realistic motion sequences. Each generated sequence is refined using the MA optimizer, which employs Reverse Learning to explore novel biomechanical patterns and Multiverse-Directing to enhance convergence, enabling optimized motion trajectories for energy efficiency, joint stability, and tactical effectiveness. The proposed TC-GAN-MA framework successfully generated high-quality, temporally coherent motion sequences that closely replicate real athlete performance. Experimental evaluation demonstrated an overall performance above 98% in both accuracy and F1-score, validating the superiority and robustness of the proposed system. The TC-GAN-MA module thus provides a robust and intelligent system for the simulation and enhancement of Wushu Sanshou techniques.