Research on Tennis Motion Evaluation Method and System Based on Deep Learning

Abstract

The current tennis training lacks the guidance to standardize the user action evaluation method; it is difficult to use the human movement analysis related technology to evaluate the user action and feedback. In this paper, a real-time evaluation algorithm for human movements in tennis training with the monocular camera is studied. Aiming at solving the problem of making use of temporal and spatial information of behavior sequence, an improved preprocessing algorithm based on depth image and bone data was proposed, and bone joint feature vectors were constructed to describe the topological shape of the human skeleton. A five-channel convolutional neural network (5C-CNN) model is proposed to effectively train multimodal behavior data. In order to solve the problem of human motion direction, the counterclockwise rotation angle of limbs was proposed as a motion evaluation feature. The experimental test on the MSR Action 3D database shows that the accuracy of the method for tennis movement evaluation reaches 98.1%, which verifies the validity of the model construction.