基于卷积神经网络的运动辅助决策系统设计  

Design of sports assistant decision system based on convolutional neural network

作　　者：查毅[1] 冯宏伟[2] ZHA Yi;FENG Hongwei(Xi’an University,Xi’an 710065,China;Northwest University,Xi’an 710069,China)

机构地区：[1]西安文理学院,陕西西安710065 [2]西北大学,陕西西安710069

出　　处：《电子设计工程》2021年第15期111-114,119,共5页Electronic Design Engineering

摘　　要：针对现有运动训练辅助决策系统智能性差、综合精度低的问题,文中基于神经网络技术设计了一套实时运动辅助决策系统。为解决常用训练数据冗余多、鲁棒性较弱的问题,通过使用OpenPose建立了人体数据实时检测与记录模块,用于实现基于视觉相机的运动姿势采集。为了对人体运动状态数据进行准确分析,采用卷积神经网络建立通用的优化模型,并对网络模型各个部分进行重新设计,实现了人体运动姿势分析识别系统。通过对实验数据的统计与分析表明,文中所提的改进系统具有训练收敛快、动作姿势识别精度高的特点,能够提供更加精准的运动辅助决策。Aiming at the problems of poor intelligence and low comprehensive accuracy of the existing decision support system for sports training.In this paper,a real-time motion aided decision-making system is designed based on neural network technology.In order to solve the problem of redundancy and weak robustness of common training data,a real-time detection and recording module of human body data is established by using OpenPose,which is used to realize the fine real-time acquisition of human motion posture based on visual camera.In order to analyze the human motion state data accurately,a general optimization model is established by using convolution neural network,and each part of the network model is redesigned,and the analysis and recognition system of human motion posture is implemented.The statistics and analysis of experimental data show that the improved system has the characteristics of fast training convergence and high accuracy of gesture recognition,which can provide more accurate motion decision-making.

关 键 词：卷积神经网络 运动辅助决策 视觉处理 状态分析 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]