基于改进YOLOv5算法的升降机人员不安全行为识别方法  

Unsafe Behavior Recognition Method of Lift Personnel Based on Improved YOLOv5 Algorithm

作　　者：余益鸿 周传德[1] 孟明辉[1] 朱志强 付朝毅 张鑫 YU Yihong;ZHOU Chuande;MENG Minghui;ZHU Zhiqiang;FU Zhaoyi;ZHANG Xin(College of Mechanical and Power Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)

机构地区：[1]重庆科技学院机械与动力工程学院,重庆401331

出　　处：《重庆科技学院学报（自然科学版）》2022年第2期79-83,98,共6页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：重庆市科技重大主题专项重点研发项目“基于视觉测量和力度感知的智能手臂关键技术及产业化”(CSTC2018JSZX-CYZTZXX0026);重庆科技学院科技创新项目“基于视觉识别的升降电梯安全行为检测系统”(YKJCX2020306)。

摘　　要：为研究施工升降机人员超载、非内部人员闯入等不安全行为,设计了基于深度学习人员数量检测和非内部人员识别的软件系统。通过增大感受野、加深CSP2网络层数、引入形变卷积、Retinex图像增强对YOLOv5检测模型进行优化。对比分析YOLOv3、YOLOv4、YOLOv5、改进YOLOv5算法在不同光照强度下的平均检测准确率和鲁棒性。实验结果显示,改进YOLOv5算法的准确率和鲁棒性有较大幅度提升,其中,超员检测正确率为100%,非内部人员检测正确率为95%。In order to study the unsafe behaviors in operating construction elevators such as personnel overload and non-internal personnel intrusion,a software system based on deep learning is designed.The YOLOv5 detection model is optimized by increasing the receptive field,deepening the CSP2 network hierarchy,introducing deformation convolution and Retinex image enhancement.The average detection accuracy and robustness of algorithms YOLOv3,YOLOv4,YOLOv5 and improved YOLOv5 are compared in different light intensity.The results show that the accuracy and robustness of improved YOLOv5 algorithm is greatly improved.Among them,the accuracy rate of overload detection is 100%,and the accuracy rate of non-internal personnel detection is 95%.

关 键 词：施工升降机 不安全行为 人员检测 改进YOLOv5算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]