基于边缘智能设备的滚装电梯载客异常检测方法及实验系统  

An Elevator Passenger Abnormality Detection Method and Experimental System Based on Edge Intelligence

作　　者：汪猛 董利达[1] 董文[1] 张慧熙[1] WANG Meng;DONG Lida;DONG Wen;ZHANG Huixi(School of Information Science and Technology,Hangzhou Normal University,Hangzhou 311121,China)

机构地区：[1]杭州师范大学信息科学与技术学院,浙江杭州311121

出　　处：《杭州师范大学学报（自然科学版）》2024年第4期439-445,共7页Journal of Hangzhou Normal University(Natural Science Edition)

基　　金：安徽省自然科学基金(2108085QH370);皖南医学院校中青年科研基金(wk202008).

摘　　要：目前常见的基于服务器架构的滚装电梯载客异常检测系统存在通信时延且成本较高问题,本文提出了一种基于边缘智能的滚装电梯载客异常检测方法.该方法采用YOLO v3推理速度优化后的网络YOLO v3-R作为目标检测算法,具体工作内容分为离线与在线2部分:离线部分包括硬件选型、YOLO v3-R网络训练与部署等;在线部分包括视频采集、载客异常目标检测与判决.在此基础上,制作了基于Jetson Nano硬件平台的滚装电梯载客异常检测实验系统,该系统无需网络通信、可靠性高、成本低,实时推理速度可达到1.96帧/s.At present,the common server architecture based ro-ro elevator passenger anomaly detection system has the problem of communication delay and high cost.This paper proposes a method of ro-ro elevator passenger anomaly detection based on edge intelligence.This method uses YOLO v3-R network optimized with YOLO v3 inference speed as the target detection algorithm.The specific work content is divided into offline and online parts.The offline part includes hardware selection,YOLO v3-R network training and deployment,etc.The online part includes video collection,abnormal passenger detection and judgment.On this basis,a ro-ro elevator abnormal detection experimental system based on Jetson Nano hardware platform is developed.The system does not need network communication,has high reliability,low cost,and real-time reasoning speed can reach 1.96 frames per second.

关 键 词：边缘智能 Roofline模型 YOLO 电梯安全 异常检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]