Research on action matching of skeletal point coordinates and sports teaching application based on Open-pose

This study addresses the challenges of high matching errors and low recognition rates in traditional skeletal point-based human action matching methods, a skeleton point coordinate and human posture action matching technology is studied based on Open-pose open-source model. Based on the Open-pose open source model, we construct a skeletal point coordinate action matching network model, use the feed-forward network for 2D confidence mapping, test it through the loss function, calculate the shortest distance to identify the association affinity domain, and introduce the greedy relaxation algorithm to optimize the accuracy rate of the association matching of multi-body skeletal points; we obtain the skeletal point coordinate parameters through the two-dimensional spatial mapping and use the k-means algorithm to quantify the features of the skeletal point coordinates, and the residuals of the skeletal point coordinates are quantized. The k-means algorithm is used to quantize the features of the skeletal point coordinates and the residual operations are concatenated to obtain the skeletal point feature vectors, which are probabilistically weighted to improve the accuracy of matching the skeletal point coordinates with the postural movements. Experimental results show that the loss rate of the skeletal point coordinate action matching method proposed in this paper is 12.4 %, and the Rank-n recognition rate can reach up to 95.2 %, which has higher accuracy, and the application in physical education is conducive to improving the completion of the course tasks, promoting the optimization of the education model, and has good application value.