Study On The Recognition Of Badminton Technical Movement Based On Convolutional Neural Network

The advances in human social civilization and the development of The Times have brought about great changes in people’s lifestyles.While taking advantage of superior material conditions,human beings face more serious health issues.As a simple and easy to use sport,badminton is not restricted by the venue and time.It has a wide audience and a large number of fans,so it is a sport suitable for people to exercise.The current research hotspot is based on action recognition technology based on convolutional neural networks,but the main work reflected in visual interaction,for the badminton action recognition and evaluation research is few,due to the badminton action complexity of higher and faster,in the process of movement is vulnerable to clothing,racket shade,so it is difficult to without Kinect depth camera,VR3 d capture equipment extract the key points of human bones to identify the action.This paper identifies the basic technical movements of badminton based on the convolutional neural network,First,the Open Pose human pose estimation model was used to obtain the key points of human bones,Secondly,the technical movements are identified by the training model,Then,the dynamic time normalization algorithm is used to evaluate the technical action,Finally,the above parts are applied to the badminton technical action evaluation system,It provides badminton fans with measures that are not restricted by time and field,At the same time for the learning and improvement of badminton technical movements,Making badminton sports become more scientific,At the same time,it also provides application cases and expand ideas for other sports.The main work of this paper is as follows:1、Summary and analysis of relevant theories and techniques.Having read a large body of relevant literature via the Internet and library,the purpose of this paper is to analyze and summarize the research status of badminton education and training,the research status of deep learning and the research progress of deep learning in the field of sports action recognition.The theories and methods of the related fields are deeply studied,including deep learning technology,convolutional neural network and human posture recognition technology.2 、 We collected the data set of experimental action samples and performed image preprocessing of the data set.First of all,this paper explains and introduces the two types of basic technical movements studied in this paper.Secondly,in the experiment,the data acquisition device is built through the mobile phone,bracket,camera and other equipment.In this paper,a total of 20 participants(There are 10 professional athletes and all are national first-class badminton players and 10 amateur badminton fans respectively)were collected for experiments,and each person collected the video data of the basic technical movements of forehand ball and forehand ball as the data set.The collected data sets are sorted out and processed,including video clips,box selection,generating continuous action frames,and image pre-processing operations such as data enhancement and expansion.Finally,the network structure of Open Pose model is introduced.We use the improved Open Pose human pose estimation model to identify key human bone points and extract features from the entire data set,including deleting redundant key points,establishing two-dimensional coordinate system,and supplementing missing key points.3、Construction and training of the identification model.Through further research and exploration of the convolutional theory of neural networks,the three mainstream convolutional neural network architectures of Res Net,VGG and Goog Le Net are analyzed.The recognition effect of these three convolutional neural network architecture on badminton technology is analyzed.The final experimental results show that the performance of the Res Net-50 model is optimal in the application scenarios studied here.4、Using dynamic time tuning algorithm of tester and professional athletes technology analysis,will match good action sequence by the euclidean distance calculation similarity between bone points,and to score the action,and mark the wrong action corresponding bone points,finally the results feedback to the tester to scientific and accurate evaluation of technical action.5、Basically realized the functions in the framework of the badminton technical action evaluation system.On the basis of the above research,the badminton technical action evaluation system is designed,which provides a platform for badminton beginners and college students to learn the technology as a technical guidance or teaching auxiliary tool.First,once the feasibility,function,and operational requirements of the system design have been analyzed,action evaluation module and visualization module.Second,all four modules are developed,and finally,the evaluation system is tested and displayed.Experimental results proved that the system can help users to correct technical action,meet the demand of scientific exercise,at the same time for the sports classroom due to limited resources of teachers is difficult to improve all the students training condition,can meet the teachers timely grasp the needs of students’ technical level,so as to improve the disadvantages of traditional teaching mode,improve the teaching quality.