基于改进YOLOv8n的道路安全头盔佩戴检测的研究  

Research on Wear Detection of Road Safety Helmets Based on Improved YOLOv8n

作　　者：陈思港 吕强 CHEN Sigang;LV Qiang(School of Electronic and Electrical Engineering,Wuhan Polytechnic University,Wuhan,Hubei 430023,China)

机构地区：[1]武汉轻工大学电气与电子工程学院,湖北武汉430023

出　　处：《自动化应用》2025年第6期44-48,共5页Automation Application

摘　　要：近年来,电动车驾驶人未佩戴头盔导致的交通事故频繁发生,造成了严重的人身伤害和财产损失。调查结果表明,事故主要集中在交通路口,因此,交通路口电动车驾驶人员头盔佩戴行为的检测与管控亟待展开。在利用机器视觉传感器获取大量电动车、驾驶人以及车牌数据后,制作合适的数据集,并在Pytorch框架上利用改进的YOLOv8n神经网络对处理后的数据进行训练,以获得最优权重参数。实验对比发现,改进后的YOLOv8n算法对于电动车、头盔以及车牌的检测精度分别达到了99.1%、93.7%和95.9%,比原始YOLOv8n的识别准确度提高了1%3%。最后,通过联合使用训练改进的YOLOv8n模型,在检测头盔佩戴情况的同时,结合电动车本身以及车牌的综合信息,准确对其进行跟踪,从而实现对交通路口违规电动车驾驶行为的有效管控。In recent years,traffic accidents caused by electric vehicle drivers not wearing helmets have occurred frequently,resulting in serious personal injuries and property damage.The survey results show that the accidents are mainly concentrated at traffic intersections,therefore,the detection and control of helmet wearing behaviour of EV drivers at traffic intersections need to be carried out urgently.After obtaining a large amount of EV,driver,and licence plate data using machine vision sensors,a suitable dataset is produced and the processed dataset is trained using the improved YOLOv8n neural network on the Pytorch framework to obtain the optimal weighting parameters.Experimental comparisons reveal that the improved YOLOv8n algorithm achieves 99.1%,93.7%and 95.9%detection accuracy for electric vehicles,helmets,and licence plates,respectively,which is 1%to 3%higher than the recognition accuracy of the original YOLOv8n.Finally,through the joint use of the training improved YOLOv8n model,while detecting the wearing of helmets,combined with the comprehensive information of the EV itself and the licence plate,the EV is accurately tracked,so as to achieve effective control of illegal EV driving behaviour at traffic intersections.

关 键 词：神经网络 目标检测 YOLOv8n 头盔 轻量化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]