Research on action recognition for college dance teaching based on deep learning

Abstract

Abstract In this paper, Gaussian modeling is established to perform background subtraction of action feature images of dancers and combined with median filtering to denoise the background of feature images. The cumulative edge feature algorithm extracts relevant features from the dance dataset using the dance itself. Based on this, the directional gradient is introduced to output the extracted dance histogram features and cascade the optical flow histogram feature vectors in all blocks to form the HOF features of the image. The deep learning model is used to design the dance teaching model, and the teaching objectives are determined. The effect of dance teaching is tested through the method of simulation experiments and empirical analysis. The loss value of the model without deep learning is very high, reaching 1.25, and the loss values after convergence of the parametric model using deep learning are 0.85 and 0.95, and the deep learning algorithm has a better optimization effect on the model. The scores of the seven measures of teaching effectiveness are all above 12, and the significance p-value of each indicator is <0.05. The dance teaching effect is significant.