Dance Movement Recognition Method Based on ST-GCN and Attention Mechanism - A Study Combining Athlete Physical Training Assistance

This study introduces a novel approach for precise dance action recognition through the integration of an attention mechanism and spatio-temporal graph convolutional network (ST-GCN) with athlete physical training assistance. The ST-GCN serves as the foundational framework for recognizing dance actions, augmented by an attention mechanism to mitigate the limitations of fixed topology structures. This addition enables the system to concentrate on crucial features during dance action recognition training, thereby enhancing the model's recognition efficacy. Additionally, an adaptive mechanism is incorporated into the network to refine the training of ST-GCN. The efficacy of the proposed method is evaluated utilizing a dataset of human (skeleton) behaviors containing 56,880 samples, partitioned into training and testing sets at a ratio of 2:1. Results demonstrate that the proposed method surpasses conventional deep neural network approaches for action recognition, achieving an average accuracy of 94.90% on the test set, representing a 29.08% improvement. This study contributes to the application of deep learning and artificial intelligence in sports and enhances dance pedagogy by providing coaches and dancers with a deeper understanding and mastery of intricate movements. Moreover, the method holds promise for applications in fitness and sports competitions.