三维机器视觉在自动扶梯乘客危险行为识别与预警的应用  

Application of 3D Machine Vision in Escalator Passenger Dangerous Behavior Recognition and Early Warning

作　　者：欧阳惠卿[1] 舒文华[1] 李杨[2] 祝瑞男 张瑞 Ouyang Huiqing;Shu Wenhua;Li Yang;Zhu Ruinan;Zhang Rui(Shanghai Institute of Special Equipment Inspection and Technical Research,Shanghai 200062,China;School of Electronic Science and Engineering,Nanjing University,Nanjing 210023,China;Beijing Metro Operation Co.,Ltd.,Electromechanical Branch,Beijing 100044,China;Beijing Metro Technology Development Co.,Ltd.,Beijing 100044,China)

机构地区：[1]上海市特种设备监督检验技术研究院,上海200062 [2]南京大学电子科学与工程学院,南京210023 [3]北京市地铁运营有限公司机电分公司,北京100044 [4]北京地铁科技发展有限公司,北京100044

出　　处：《机电工程技术》2021年第11期59-62,共4页Mechanical & Electrical Engineering Technology

基　　金：国家市场监管总局科研项目(编号:2019MK019)。

摘　　要：为了降低因为乘客危险行为导致的自动扶梯安全事故/伤人事件,基于三维视觉和人工智能技术,提出了一种自动扶梯乘客危险行为识别和预警技术,并详细介绍了该系统主要硬件和现场布置方案。为了对该系统进行科学评价,还提出了准确率、查全率、误报率、响应时间等综合性能评价指标,并分别在实际现场和实验室模拟条件下对该系统进行了大量的测试,测试结果表明,所构建的系统在不同光照、不同客流强度环境下,均能有效识别自动扶梯乘客携带婴儿车、摔倒、逆行、出入口逗留、出入口拥堵、探头等多种危险行为。In order to reduce the escalator safety accidents/injuries caused by passengers′dangerous behaviors, based on three-dimensional vision and artificial intelligence technology, a kind of escalator passengers′dangerous behavior recognition and early warning technology was proposed, and the main hardware and layout scheme of the system was introduced in detail. In order to evaluate the system scientifically, also the comprehensive performance evaluation indexes were put forward, such as precision, recall, false alarm rate, average response time, etc., and a lot of tests under the actual field and laboratory simulation conditions were carried out respectively. The test results show that the system constructed has good performance under different lighting and different passenger flow intensity environments. It can effectively identify a variety of dangerous behaviors of escalator passengers, such as carrying baby carriage, falling, retrograde, staying at the entrance or exit,congestion, head out of handrail and so on.

关 键 词：自动扶梯 乘客行为 三维视觉 

分 类 号：TU857[建筑科学]