Research On Table Tennis Tactics Data Automatic Detection Based On Deep Neural Network

With the development of computer technology,especially the widespread application of computer vision technology,statistical analysis techniques for digital images and video are becoming more and more important,especially for high-level semantic analysis of video data(such as the type of objects present in the video,the motion of the analyzed object,etc.)which is becoming a hot research area.One of the important applications is the analysis of sports videos.Just like the audience watching sports games,people hope that computers can understand the process of sports competitions.This not only assists athletes and coaches in post-match analysis,but also designs robot-assisted teaching training.The same is true for table tennis matches.The purpose of this paper is to design and implement a set of sports system for table tennis and automatic detection of technical and tactical indicators.The main technical and tactical indicators include motion trajectory and rotation information.In order to achieve this goal,the main contents and contributions of this thesis include: the detection and tracking algorithm of table tennis based on deep neural network is proposed,which achieves better results in self-built video dataset and actual system testing.After obtaining the position of table tennis in the image,this paper proposes a single-view point table tennis depth recovery algorithm to avoid the difficulties of hardware synchronization encountered in multi-view video shooting.Then,in order to calculate the rotation direction and speed of table tennis,this paper studies and implements the rotation measurement method based on Fourier transform and the rotation direction detection method based on CNN,which achieves better results in the lower speed data set test.Finally,this paper proposes a trajectory prediction algorithm based on LSTM,which lays a foundation for the design of table tennis robot by predicting the ping-pong trajectory.The final experimental test shows that the system implemented in this paper can better deal with the problem of table tennis tracking and rotation measurement.