基于OR-CNN的电动车进入电梯危险行为检测系统设计  

Design of Detection System of Dangerous Behavior of Electric Vehicle Entering Elevator Based on Deep Learning

作　　者：吕樵润 林辉[1] 刘孝炜 LYU Qiaorun;Lin Hui;Liu Xiaowei(School of Intelligent Engineering,Shaoguan University,Shaoguan,Guangdong 512005,China)

机构地区：[1]韶关学院智能工程学院,广东韶关512005

出　　处：《机电工程技术》2024年第1期253-256,共4页Mechanical & Electrical Engineering Technology

基　　金：2022年广东省大学生创新创业训练项目资助(S202210576069)。

摘　　要：在楼宇中电动车不论是在电梯中还是在楼层中自燃爆炸都会给人们造成严重的危害,是一个重大的危险源。目前虽然在电梯口贴上了“电动车禁止进入”的提示标志,但效果不佳。因此,设计了基于OR-CNN的电动车进入电梯危险行为检测系统,以满足管理部门对禁止电动车进入电梯的需求。基于OR-CNN网络的电动车检测模型将RoI池化层替换为PORoI,PORoI池化单元通过先验知识将目标划分为5个部分,融合各个部分的特征信息,更好地完成在遮挡环境下的目标检测任务。此外,系统在发现违规行为时会使电梯门处于禁关状态,并发出警报提醒,对违规行为的视频段进行抽帧处理并记录存档,以便事后追责,实现智能化管理。测试结果表明,与YOLOv5相比,所设计的检测系统在遮挡情况下的电动车检测准确率明显提高,更适应电梯等狭小环境中目标遮挡的情况。In buildings,whether it is in an elevator or in a floor,the spontaneous combustion and explosion of electric vehicles will cause serious harm to people and is a major source of danger.At present,although the elevator entrance is posted with a reminder sign that"electric vehicles are not allowed to enter",the effect is not good.Therefore,an OR-CNN based detection system for the dangerous behavior of electric vehicles entering the elevator is designed to meet the needs of the management department for prohibiting electric vehicles from entering the elevator.The electric vehicle detection model based on OR-CNN network replaces the RoI pooling layer with PORoI,and the PORoI pooling unit divides the target into five parts through prior knowledge,fuses the feature information of each part,and better completes the target detection task in the occlusion environment.In addition,when the system finds a violation,the elevator door will be in a forbidden-to-close state,and send an alarm in the corresponding elevator,the video segment of the violation is recorded and archived,so as to hold accountable afterwards and achieve intelligent management.The test results show that compared with YOLOv5,the electric vehicle detection accuracy of the designed detection system is significantly improved in the case of occlusion,and it is more suitable for target occlusion in narrow environments such as elevators.

关 键 词：OR-CNN PORoI 电动车 电梯 危险行为检测 特征信息融合 管理智能化 

分 类 号：TU857[建筑科学] TP391.41[自动化与计算机技术—计算机应用技术]