面向智能体能检测系统的三维姿态估计方法研究及应用  

Research and Application of 3D Attitude Estimation Method for Intelligent Physical Fitness Detection System

作　　者：刘立业 盖璇 LIU Liye;GAI Xuan(Northeast Petroleum University,Daqing,Heilongjiang 163000,China)

机构地区：[1]东北石油大学,黑龙江大庆163000

出　　处：《自动化与仪器仪表》2023年第12期197-200,共4页Automation & Instrumentation

基　　金：黑龙江省重点课题《现阶段"三亿人上冰雪"活动在黑龙江省高校深入推进的问题与策略研究》(GJB1422176)。

摘　　要：为了提升体能检测中的计数效率,研究提出一种结合堆叠沙漏模型(Stacked Hourglass Model,Hourglass)和超轻量级神经网络的三维人体姿态估计模型,并引入了K邻近算法(K Nearest Neighbors,KNN)来实现动作分类与计数。实验结果显示,所提模型的计数精度可达99.66%,实际测试的帧速在27.41 FPS左右。对不同体能姿态估计的平均召回率在99.33%。与传统的计数方式相比,本智能计数模型具有更高的计算精度和更低的成本。由此,本次实验所构建的模型具有较强的适用性。To improve the counting efficiency in physical fitness detection,a three-dimensional human pose estimation model combining Stacked Hourglass Model(Hourglass) and ultra lightweight neural network is proposed,and the K Nearest Neighbors(KNN) algorithm is introduced to achieve action classification and counting.The experimental results show that the counting accuracy of the proposed model can reach 99.66%,and the actual tested frame rate is around 27.41 FPS.The average recall rate for estimating different physical postures is 99.33%.Compared with traditional counting methods,this intelligent counting model has higher computational accuracy and lower cost.Therefore,the model constructed in this experiment has strong applicability.

关 键 词：三维姿态估计 体能计数 堆叠沙漏模型 超轻量级神经网络 KNN 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TH89[自动化与计算机技术—计算机科学与技术]