基于YOLOv5的行人口罩佩戴实时检测  

Real-time Detection of Pedestrian Mask-wearing based on YOLOv5

作　　者：张熙 庞爱民 ZHANG Xi;PANG Ai-Min(School of Mechanical Engineering and Autonation,Wuhan Textile University,Wuhan Hubei 430200,China)

机构地区：[1]武汉纺织大学机械工程与自动化学院,湖北武汉430200

出　　处：《武汉纺织大学学报》2023年第2期41-44,共4页Journal of Wuhan Textile University

摘　　要：针对新冠疫情期间人工检查行人口罩佩戴情况效率低下的问题,提出了基于YOLOv5网络来实现对行人口罩佩戴情况的实时检测算法。收集了2000张佩戴口罩及未佩戴口罩行人图片作为数据集,先基于COCO数据集的权重数据进行预训练,提高训练的速度和检测;再将数据集导入YOLOv5模型中进行迭代训练及测试,将所获得的最优权重文件对测试集进行验证,并把训练结果可视化展示。实验结果表明,该算法在行人密集的情况下实时检测速度也能达到62.5FPS的高准确率,满足了行人口罩佩戴实时检测的要求。To address the inefficiency of the current mainstream manual inspection of pedestrian mask wear,a YOLOv5 network-based approach is proposed to achieve real-time detection of pedestrian mask wear.To address the inefficiency of the current mainstream manual detection of pedestrian mask wear,a YOLOv5 network-based algorithm is proposed to achieve real-time detection of pedestrian mask wear.2000 images of masked and unmasked pedestrians were collected as datasets,and pre-training was first performed on weighted data based on the COCO dataset to improve the speed and detection of training.Then,the datasets are imported into the YOLOv5 model for iterative training and testing,the obtained optimal weight files are validated against the test sets,and the training results are presented visually.Experimental results show that the algorithm can exceed 62.5FPS in real-time detection even when pedestrians are dense,with high accuracy,meeting the requirements for real-time detection of pedestrian mask wearing.

关 键 词：YOLOv5 实时检测 COCO 最优权重 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]