基于YOLOv5的人员安全帽检测告警算法

Personnel Safety Helmet Detection and Alarm Algorithm Based on YOLOv5

佩戴安全帽可有效保障作业人员安全,避免事故的发生,然而安全帽佩戴的实时监控问题却一直未得到良好的解决。为了缓解电力场所安全监控压力,基于YOLOv5目标检测算法设计了一套人员安全帽佩戴检测和危险区域告警算法,通过对锚框进行重聚类并引入注意力机制等对原始网络进行改进,结合公开数据集和自建数据集对网络进行训练,检测精度达到89.9%。最后通过Jetson Xavier NX等硬件设备将该算法部署到机器人上进行巡检,可实现对作业现场人员的实时监管,有效保障作业现场人员安全。

Wearing helmet can effectively ensure the safety of operators and avoid accidents.However,the problem of real-time monitoring of wearing helmets has not been well solved.In order to alleviate the pressure of safety monitoring of power sites,a set of safety helmet wearing detection and dangerous area alarm algorithms are designed based on YOLOv5 target detection algorithm.The original network was improved by clustering anchor frames and introducing attention mechanisms.The network is trained using both public and self-built datasets,with a detection accuracy of 89.9%.Finally,the algorithm is deployed to the robot for inspection through hardware devices such as Jetson Xavier NX,which can realize real-time supervision of job site personnel and effectively ensure the safety of job site personnel.