基于姿态估计的粮仓作业人员摔倒检测研究  

Falling Detection Study on Granary Workers Based on Human Pose Estimation

作　　者：孙福艳[1] 桂崇文 吕宗旺[1] 甄彤[1] SUN Fu-yan;GUI Chong-wen;LYU Zong-wang;ZHEN Tong(College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001,China)

机构地区：[1]河南工业大学信息科学与工程学院,河南郑州450001

出　　处：《计算机仿真》2024年第5期236-241,共6页Computer Simulation

基　　金：国家科技支撑计划(2018YFD0401404)。

摘　　要：近些年时常发生粮仓作业人员因意外跌入粮堆无法自救、熏蒸作业中毒晕倒等安全问题。为了及时获取有效信息,减少对工人的伤害,提出一种自下而上的人体姿态估计方法,建立对粮仓工作人员状态检测模型。首先制作不同角度的粮仓作业视频和摔倒状态公共数据集;接着改进轻量级OpenPose算法,数据集引入算法生成骨骼图像;最后设计摔倒检测算法,得到一种基于姿态估计的粮仓作业人员摔倒检测模型。实验结果表明,检测摔倒状态准确率有所提高,轻量级OpenPose算法计算量较少,内存占用小,可以实时地检测粮仓作业人员摔倒状态。In recent years,there have been frequent safety issues such as the inability of grain warehouse workers to self rescue due to accidental falls into grain piles,and fainting due to poisoning during fumigation operations.In order to timely obtain effective information and reduce harm to workers,a bottom-up human pose estimation method is proposed,and a state detection model for granary workers is established.Firstly,we produced a public dataset of grain bin operation videos and falling states from different angles;Then we improved the lightweight OpenPose algorithm and introduced the dataset into the algorithm to generate skeletal images;Finally,we designed a falling detection algorithm to obtain a fall detection model for grain bin operators based on pose estimation.The experimental results show that the accuracy of detecting fall state is improved,and the lightweight OpenPose algorithm has lower computational complixity and small memory occupation,and can detect the fall state of grain bin operators in real-time.

关 键 词：人体姿态估计 摔倒检测 轻量级算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]