Action Recognition Method for Dance Teaching Video by Integrating Graph Convolutional Network and Attention Mechanism

The utilization of artificial intelligence technology in the area of dance action recognition is receiving increasing attention. However, traditional dance action recognition methods rely heavily on manually designed feature extraction, which makes it difficult to accurately capture the subtle differences in complex and varied dance movements, such as posture changes and rhythm changes. Based on this, a dance teaching video action recognition model based on a graph convolutional network is proposed. An attention mechanism is incorporated to enhance its performance, enabling it to perform feature processing in the temporal dimension. To confirm the capability of the model, a comparative experimental method was used to test it on the public dataset HMDB51 and a self-built dance teaching video dataset. The outcomes indicated that the recognition accuracy of the proposed model on the HMDB51 dataset reached 94.6%, and the recognition accuracy on the self-built dance teaching video dataset was $\mathbf{9 3. 8 \%}$. Through confusion matrix analysis, it was further found that the model excellent exhibited performance in the recognition tasks of five dance movements, including ballet, hip-hop, Latin, tango, and jazz, significantly outperforming other state-of-the-art comparative models. Overall, the results demonstrate that the proposed model can effectively recognize movements in dance teaching videos.