Resnet-Based Intelligent Recognition Algorithm and Evaluation of Students' Tennis Movement in Teaching Video

Abstract

In order to accurately identify and evaluate tennis movement, a method of tennis movement recognition and evaluation based on ResNet is proposed by combining computer vision with tennis movement-related knowledge. Firstly, the pose estimation model is used to detect the human target in a tennis video and extract the key points of the skeleton. Then, the ResNet model is trained using the video data set collected on the professional tennis court. The model can classify the sub-actions of tennis. A dynamic time-warping algorithm is used to evaluate the classified actions. A large number of experiments are carried out based on the collected video data set. The results show that the accuracy of the proposed ResNet-based tennis motion recognition method for the classification of 6 types of tennis sub-movements can reach 90.8%. Compared with methods based on graph convolution networks such as AGCN and ST-GCN, it has a stronger generalization ability. The proposed scoring algorithm based on dynamic time regulation gives the evaluation scores of corresponding actions in real time and accurately after the action classification, thus reducing the work intensity of tennis teachers and effectively improving the quality of tennis teaching.