Research on Dance-Specific Action Recognition Method Based on Spatio-Temporal Skeleton Graph
Yu Wang, Jiaqin Gong · 2024
With developments made at breakneck speeds in artificial intelligence and computer vision, recognizing dance movement has become a focal point of research in the realm of intelligent entertainment and art education. However, due to the intricate spatio-temporal properties of body movements and obstacles in the particular action detection, the available methods cannot hit the nail on the head in the areas of accuracy and generalization. This paper introduces a method of identifying a specific dance move based on the spatio-temporal skeleton graph. Through the process of collecting the skeleton data of dancers and creating the spatio-temporal skeleton graph, the method encodes the spatio-temporal movement characteristics of the dancers at the graph structure so as to effectively recognize the specific actions. In this study, we're looking at data preprocessing, feature extraction, classification model design, and optimization strategies. The experimental results show that this method surpasses the conventional ones in terms of recognition accuracy and robustness and therefore it can be used successfully in dance analysis and applications. These results of the proposed study are likely to be adopted widely in dance education, art analysis, and cultural heritage preservation.