A Sports Video Analysis System Based On Detection Of Human Key Points

In recent years,all countries in the world have regarded the performance of competitive sports as an important indicator for assessing the comprehensive strength of a country.Facing the increasingly fierce competition in international competitive sports,we must rely on more scientific and efficient training methods to improve the level of competitive sports.At present,the use of motion sensors to collect motion data has the disadvantages of being easily disturbed by the environment,and the accuracy of athletes’ movements through manual analysis is not high.In response to the above problems,this thesis uses deep learning technology break through the current limitations of relying on sensors to collect athletes’ sports data,and designs and implements a sports video analysis system based on the detection of key points of the human body.This system aims to monitor and analyze athletes’ movements in videos,and develop targeted training programs for coaches.Provide reference to improve the training efficiency of athletes.The system also supports effective management of system users,training teams,and training data.The main work of the thesis is as follows:(1)Demand and feasibility analysis of sports video analysis system.This thesis sorts out the overall functions of the system according to the construction goal of the sports video analysis system.In response to the requirements for high-accuracy monitoring of athletes’ movements and efficient management system training data,six functional modules such as movement analysis and training record management are designed,and the business functions and interaction logic of each module are elaborated.In addition,this thesis combines the actual application scenarios of the sports video analysis system to optimize the human body key point detection algorithm(OpenPose).Construct a data set(SV-Pose)using the weightlifting training video of athletes from the Shaanxi Provincial Sports Bureau.And finish the performance test of the OpenPose algorithm on the dataset to verify the feasibility of the algorithm technical scheme in the system.(2)Design and implementation of sports video analysis system.In terms of system architecture,this thesis builds a sports video subsystem based on the B/S architecture,uses SpringBoot technology to build the overall system back-end architecture,uses Vue.js technology framework and combines Element UI and ECharts components to visualize system data.In terms of athlete motion analysis,the OpenPose algorithm is used to detect the key points of the athlete’s bones in the video.Based on the continuity of the video frame,the key point data of the bones is optimized and the athlete’s motion parameters are calculated.In terms of data persistence,the relational database MySQL and file database are used to store system user,organization information,and sports video data.The system is designed based on loosely coupled architecture to ensure the flexibility and scalability of the system.(3)Sports video analysis system function and performance test.After the completion of the system development,this thesis deploys the system test environment and tests the system functions and performance.For system function testing,design test cases based on the analysis results of system requirements to detect the implementation effects of each module function of the system.For system performance testing,use the Jmeter stress testing tool to monitor system response time,throughput and other performance under different concurrency,and analyze and evaluate the pros and cons of system performance.The test results show that the functions of each module of the sports video analysis system are operating normally,and the system performance meets the requirements of the partner.