Research on Modern Dance Performance Forms and Dance Movement Characteristics Based on Multi-scale Feature Fusion

Due to the complexity of dance performance forms, dance movement recognition is very difficult, and related research has received much attention. In this paper, a dance action feature recognition method based on multi-scale feature fusion is designed around modern dance performance forms and dance action features. The method carries out dance action feature extraction on the well-framed image of dance video by calculating the image entropy of optical flow map, combined with the audio feature sequence of audio feature extraction. Then the multi-core learning method is used to realize the multi-feature fusion of the dance action to complete the extraction and recognition of dance action features. Through the performance analysis of the model, the recognition accuracy of this paper’s model on the NTU RGB-D 60 and NTU RGB-D 120 datasets stays above 90% and 85%, with an overall decrease of only 4.95% in the latter, which is the best performance among the compared algorithms. Experiments show that the proposed method in this paper has better dance movement recognition effect and generalization ability.