基于视觉的工人高处攀爬不安全行为识别模型  

Unsafe behavior recognition model of high climbing workers based on vision

作　　者：张泽辉 张乾隆 徐晓滨[1] 赵祖国 王海泉[3] 李昊 ZHANG Zehui;ZHANG Qianlong;XU Xiaobin;ZHAO Zuguo;WANG Haiquan;LI Hao(China-Austria Belt and Road Joint Laboratory on Artificial Intelligence and Advanced Manufacturing,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China;Secondary Vocational Internet of Things(Hubei)Information Technology Co.,Ltd.,Wuhan Hubei 430014,China;School of Electronic Information,Zhongyuan University of Technology,Zhengzhou Henan 450007,China;Ningxia Changjun Technology Consulting Co.,Ltd.,Yinchuan Ningxia 750001,China)

机构地区：[1]杭州电子科技大学中国-奥地利人工智能与先进制造“一带一路”联合实验室,浙江杭州310018 [2]中职物联(湖北)信息科技有限公司,湖北武汉430014 [3]中原工学院电子信息学院,河南郑州450007 [4]宁夏长骏科技咨询有限公司,宁夏银川750001

出　　处：《中国安全科学学报》2025年第2期144-151,共8页China Safety Science Journal

基　　金：国家重点研发计划资助项目(2022YFE0210700);国家水运安全工程技术研究中心开放基金资助(A202403);浙江省自然科学基金资助(LTGG24F030004);浙江省“尖兵”,“领雁”研发攻关计划资助项目(2024C03254)。

摘　　要：为精准识别高处作业人员攀爬过程中出现的不安全行为,提出一种基于视觉的工人高处攀爬不安全行为识别模型,由人体姿态估计算法和一维卷积不安全行为识别算法组成。人体姿态估计研究者采用量子化自编码器对人体关键点进行结构化建模,实现人体关键点坐标的检测;结合高处作业安全行为知识,基于一维卷积神经网络(1DCNN)模型构建高处攀爬不安全行为识别算法,并通过实验进行验证。结果表明:该模型在人员无遮挡情况和有部分遮挡情况下,分别达到93.91%和90.34%的精度;与支持向量机(SVM)、K最邻近算法(KNN)相比,该模型具有更强的泛化能力。In order to accurately identify unsafe behaviors during the climbing process of high-altitude workers,this paper proposed an unsafe behavior recognizing method for high climbing workers based on vision,which included the human pose estimation and the one-dimensional convolutional unsafe behavior recognition models.Quantized autoencoder was used to structurally model human key points in human pose estimation,enabling the detection of human key point coordinates.Combining with safety behavior knowledge in high climbing operations,the unsafe behavior recognition model was constructed based on one-dimensional convolutional neural network model,and it was validated by industrial data experiments.Experimental results show that the accuracy of this method is 93.91%and 90.34%on unobstructed and partially obstructed datasets,respectively.Moreover,compared with support vector machines(SVM)and K-nearest neighbor(KNN),this method has stronger generalization capability.

关 键 词：计算机视觉 工人高处攀爬 不安全行为 识别模型 一维卷积神经网络(1DCNN) 人体姿态估计 

分 类 号：X910[环境科学与工程—安全科学]