基于机器视觉的危化品事故场景人因行为检测研究进展  

Research Progress on Human Behavior Detection in Hazardous Chemical Accident Scenes Based on Machine Vision

作　　者：王春 Wang Chun(State Key Laboratory of Chemical Safety,Shandong,Qingdao,266104;SINOPEC Research Institute of Safety Engineering Co.,Ltd.,Shandong,Qingdao,266104)

机构地区：[1]化学品安全全国重点实验室,山东青岛266104 [2]中石化安全工程研究院有限公司,山东青岛266104

出　　处：《安全、健康和环境》2024年第12期1-8,共8页Safety Health & Environment

基　　金：国家重点研发计划(2023YFC3011100),面向大型复杂火场的举高喷射消防机器人关键技术研究及应用。

摘　　要：危险化学品行业事故因所涉物料多易燃易爆、生产工艺高温高压等特性,当事故发生时,事故应急处置是否及时有效,很大程度上决定了事故的后续发展,人的应急处置行为在救援过程中至关重要。近年来,大量基于机器视觉的危化品事故场景人因识别方法被提出。对现有的基于机器视觉的危化品场景人员行为检测方法进行梳理,对危化品场景人因行为检测实验数据集进行总结,力求为后续相关研究提供新的参考。介绍相关实验数据集,并通过实验比较分析现有方法的性能。首先介绍了近年来基于机器视觉的危化品场景人因行为检测方法。然后对不同方法的优缺点进行全面的总结,对其适用场景进行分析。随后,对现有的危化品事故场景人因行为检测数据集进行统计分析,并根据数据类型将检测数据集进行分类,提出了现有数据集中存在的问题及改进方法。最后,对基于机器视觉的危化品事故场景人因行为检测发展趋势进行了展望。Accidents in the hazardous chemical industry are characterized by the high flammability and explosiveness of the materials involved,as well as the high temperature and pressure of the production process.When an accident occurs,the timely and effective emergency response largely determines the subsequent development of the accident.Human emergency response behavior is crucial in the rescue process.In recent years,a large number of machine vision based methods for identifying human factors in hazardous chemical accident scenarios have been proposed.The existing machine vision based methods for detecting human behavior in hazardous chemical scenes was sorted out,and the experimental dataset of human behavior detection in hazardous chemical scenes was summarized to provide new references for subsequent related research.The relevant experimental datasets were introduced,and the performance of existing methods through experiments was compared and analyzed.Firstly,the human behavior detection methods for hazardous chemical scenes based on machine vision in recent years were introduced.Then the advantages and disadvantages of different methods were comprehensively summarized,and their applicable scenarios were analyzed.Subsequently,statistical analysis was conducted on the existing dataset of human behavior detection in hazardous chemical accident scenarios,and the detection dataset was classified according to data types.The problems and improvement methods in the existing dataset were proposed.Finally,the development trend of human behavior detection in hazardous chemical accident scenarios based on machine vision was discussed.

关 键 词：危化品事故 智能视频监控系统 人因行为 机器视觉 行为识别 行为检测方法 

分 类 号：X937[环境科学与工程—安全科学]