Design And Implementation Of Basketball Assistant Training System Based On 3D Action Recognition

In recent years,Video-based human action recognition has become one of the research hotspots in the field of computer vision,and has been widely used in fields such as intelligent human-computer interaction and virtual reality.Human action recognition is divided into 2D and 3D-based recognition methods.The 2D recognition method extracts the position characteristics of the two-dimensional plane in real time,so that the network model can only learn the position contour or color information,and it is difficult to directly reflect the nature of the movement.This is also such a method.Identify the reasons for lower intensive reading.Using the method based on 3D skeleton data information,the spatial information of the coordinate position of the joint point and the change information of the spatial position of the joint point in the time series are learned through the network model to achieve the classification effect.In order to solve the above problems,this paper proposes a fusion method of 3D pose estimation and action recognition for recognition and classification.The main contributions of this article are as follows:(1)Since the currently used data sets including NTU-RGBD,Kinect-skeletons,etc.are based on human daily activities,there is no public data set related to basic basketball actions,so I will lead it,assisted by laboratory team members,and use multiple channels.Collect and process basketball sports videos,create basketball sports data sets,which can later be made publicly available for reference for other researchers.(2)Based on the basketball data,this paper proposes a method that combines human posture estimation and action recognition,which provides a new perspective on basketball assistive training.Compared with the traditional single action recognition method,this method makes the recognition The accuracy rate is better,and the display effect is more intuitive.(3)This paper studies the 3D action recognition method.First,the multi-person human body pose estimation method is used to extract 2D skeleton information from basketball basic action video data and convert it into 3D skeleton information,replacing the data with the neural network model of spatio-temporal dual flow.The input is used for action recognition and classification to improve the recognition rate of action recognition.(4)Using the method proposed in this article,we designed and implemented a set of basketball auxiliary training system,which combines the functions of algorithm analysis and data analysis.By uploading basic action videos and basketball game videos,it provides players and athletes with Basic action training,event prediction and other services help players find shortcomings and improve their scoring rate.