Automatic Recognition Method of Table Tennis Motion Trajectory Based on Deep Learning in Table Tennis Training

Abstract

In table tennis training, in view of the problem of large errors when tracking fast moving targets, this study proposes an automatic identification method of table tennis motion trajectory based on deep learning. The multi-view image of the target object is collected by a multi-eye camera and a stereo image pair is formed. After stereo matching, the three-dimensional coordinate group of the target object is obtained by using the three-dimensional positioning principle of stereo vision. In the three-dimensional coordinate system, the mathematical model of table tennis motion is established. The initial position of the table tennis ball is detected by the Vibe algorithm, and the target area frame is marked, and the frame of the detected target area is used as the first frame of the KCF target tracking to track the table tennis ball. Based on this, a rotating table tennis trajectory recognition network is constructed based on LSTM. The experimental results show that the total trajectory error of this method is 39.00 mm, which can accurately identify the motion trajectory.