大型枢纽施工区人员异常侵入检测算法研究  

Research on algorithms for detecting abnormal personnel intrusion in large⁃scale hub construction zones

作　　者：刘夏润 罗玉萍 石杰红[3] 赵晨[3] LIU Xiarun;LUO Yuping;SHI Jiehong;ZHAO Chen(China Railway Engineering Consulting Group Co.,Ltd,Beijing 100073,China;Shenzhen Metro Construction Group Co.,Ltd.,Shenzhen Guangdong 518026,China;China Academy of Safety Science and Technology,Beijing 100012,China)

机构地区：[1]中铁工程设计咨询集团有限公司,北京100073 [2]深圳地铁建设集团有限公司,广东深圳518026 [3]中国安全生产科学研究院,北京100012

出　　处：《中国安全生产科学技术》2024年第S1期218-225,共8页Journal of Safety Science and Technology

摘　　要：为了解决大型综合交通枢纽施工范围广、作业项目复杂情况下的高作业风险问题,针对既定施工危险区域,提出1种基于Yolov5模型的人员异常侵入检测及分级预警算法,以低成本、强独立性、覆盖范围广的优势降低施工作业风险。首先,基于Yolov5模型实现施工区域内的人员检测以及人员位置信息参数化;然后,根据人员位置参数化描述实现异常侵入检测以及分级预警。研究结果表明:该算法能够在人员多姿态、多人员、人员遮挡等挑战场景下实现有效的异常侵入检测及分级预警,其有效性及鲁棒性能在公开的玛珈工地(加油站)物品数据集(MaJia Construction Site(Gas Station)Dataset,MJCSD)上得到验证。研究结果可为改善施工安全、降低预警成本提供新的解决思路,具有实际工程应用价值。In order to solve the high operation risk problem of large⁃scale integrated transportation hubs with wide construction scope and complex operation projects,an abnormal intrusion detection and graded warning algorithm for personnel based on the Yolov5 model for the established construction danger zone was proposed to reduce the construction operation risk with low⁃cost,strong independence and wide coverage advantages.Firstly,based on Yolov5 model,personnel detection and parameterization of personnel position information in the construction area were realized;then abnormal intrusion detection and graded warning were realized according to the parameterized description of personnel position.The research results showed that the algorithm proposed in this paper could realize effective abnormal intrusion detection and graded warning in challenging scenarios such as multi⁃pose,multi⁃personnel and personnel occlusion,whose effectiveness and robust performance were verified on the public MaJia Construction Site(Gas Station)Dataset(MJCSD).The research results can provide new solutions for improving construction safety and reducing warning costs,and have practical engineering application value.

关 键 词：大型枢纽 异常侵入 目标检测 分级预警 Yolov5 

分 类 号：TU714[建筑科学—建筑技术科学] U115[交通运输工程]