一种基于YOLO v4和DeepSort的施工作业人员安全装备检测方法

A Personal Protective Equipment Detection Method at the Construction SiteBased on YOLO v4 and DeepSort

为了快速有效地识别施工人员是否完整且正确地佩戴安全装备,本文提出了一种基于YOLO v4算法和DeepSort算法的个人安全防护装备检测方法。首先,提出了一种基于统计的数据增强方法对获取的图像数据进行扩展,使训练模型具有更高的稳健性;其次,结合YOLO v4算法和DeepSort算法对人体和安全防护装备进行检测、跟踪;然后,分析人体框与安全防护装备框的IoU值和相对位置,判断每个工作人员是否完整且正确地佩戴安全装备;最后,显示未佩戴或佩戴错误的装备信息并进行报警。实验结果表明,本文中提出的方法平均检测精度较高,可满足工业应用的精确度要求,能够实现智能化、数字化的实时监控。

In order to quickly and effectively identify whether people are wearing safety equipment completely and correctly,this paper proposes a personal protective equipment detection method based on YOLOv4 and DeepSort.Firstly,a statistical-based data enhancement method is proposed to extend the acquired images to make the training model more robust;secondly,the YOLO v4 and DeepSort are combined to detect and track the human body and protective equipment;then,the IoU and the relative positions of the human bounding box and the safety equipment bounding boxes are calculated to judge whether each staff member wears the safety equipment completely and correctly;finally,the information of the equipment that is not worn or worn incorrectly is displayed and the system gives an alarm.The experimental results show that the mean Average Precision(mAP)of the method in this paper is high,which can meet the accuracy requirements of industrial applications and can realize intelligent and real-time monitoring.