基于深度学习的运动训练姿态智能分析方法研究  

Research on intelligent analysis method of sports training posture based on deep learning

作　　者：郑丽[1] ZHENG Li(College of Engineering and Technology,Chengdu University of Science and Technology,Leshan 614007,China)

机构地区：[1]成都理工大学工程技术学院,四川乐山614007

出　　处：《电子设计工程》2021年第10期167-171,共5页Electronic Design Engineering

基　　金：2017年四川省教育厅人文社会科学(休闲体育产业)重点研究基地课题(XXTYCY2017B08)。

摘　　要：采用技术手段对运动训练姿势的准确分析,是现代体育中提高运动员竞技水平的重要手段之一。针对利用人工智能技术可以精确地对运动训练姿态进行准确分析与预测的应用需求,文中基于深度学习设计了一套运动姿态分析预测系统。该方案使用Arduino嵌入式开发板为基础,搭载多个IMU传感器,通过搭配使用步进电机,建立了采集速度、加速度等准确的人体运动数据的系统,获得了人体运动的准确数据。为了准确识别人体运动模型,文中搭建了基于对称编码、时间尺度编码与结构编码的深度学习模型。实验结果验证了系统的实际性能,并通过与现有最新的ERD分类算法对比可知,该系统的分类误差低且实时性高,具有良好的二次开发前景。Using technical means to accurately analyze sports training postures is one of the important means to improve athletes'competitive level in modern sports.Aiming at the application needs of using artificial intelligence technology to accurately analyze and predict sports training postures,a set of sports posture analysis and prediction system is designed based on deep learning.This solution uses the Arduino embedded development board as the basis,is equipped with multiple IMU sensors,and uses a stepper motor to establish an accurate speed,acceleration and other human motion data collection system to obtain accurate human motion data.In order to accurately identify the human motion model,a deep learning model based on symmetric coding,time⁃scale coding and structural coding is built in this article.The experimental results verify the actual performance of the system,and by comparing with the latest existing ERD classification algorithm,it can be seen that the system has low classification error and high real⁃time performance,and has a good prospect for secondary development.

关 键 词：深度学习 姿态分析 混合编码 数据分类 

分 类 号：TN99[电子电信—信号与信息处理] TP393[电子电信—信息与通信工程]