摘要
针对现有体能考核中俯卧撑计数裁判负荷大、效率低的问题,设计一种基于机器视觉的俯卧撑计数算法。该算法使用改进的YOLOv3进行人体目标检测,由Ultralight-SimplePose预测出俯卧撑时人体的关键点分布,之后利用SVM分类器,训练SVM模型,将俯卧撑的几种阶段分类,进行计数。测试结果表明,该算法人体关键点识别率可以达到0.945,俯卧撑计数准确率大于99%,可以准确实现体能训练或考核中的人体俯卧撑姿态识别,并进行计数功能。
In order to solve the problems of heavy load and low efficiency of push-up counting referee in the existing physical fitness assessment,a push-up counting algorithm based on machine vision is designed.The system uses improved YOLOv3 to detect human targets,and then us⁃es Ultralight-SimplePose to predict the distribution of the key points of the human body during push-ups.After that,SVM classifier is used to train SVM model,and several stages of push-ups are classified and counted.The test results show that the recognition rate of key points of human body can reach 0.945,and the deviation of push up count is less than 1%.It can accurately realize the recognition of push-up posture in physical training or examination,and carry out the counting function.
作者
徐菲
陶青川
吴玲
敬倩
XU Fei;TAO Qingchuan;WU Ling;JING Qian(School of Electronic Information,Sichuan University,Chengdu 610065)
出处
《现代计算机》
2021年第15期48-53,共6页
Modern Computer