基于改进YOLOv5s的骑行头盔佩戴检测算法研究

Research on Cycling Helmet Wearing Detection Algorithm Based on Improved YOLOv5s LI Hong-zhi,SHU Yuan-zhong,XIAO Jing,NIE Yun

电动自行车因其出行便利性,逐渐成为主流出行方式。但是在道路交通事故中,电动车骑行人员伤亡率居高不下。为解决电动车高伤亡率的问题,采用改进的YOLOv5s对道路场景下的骑行头盔佩戴情况进行检测。首先,用GSConv Module代替原有YOLOv5s骨干网络中的标准卷积,在保证检测精度的同时,提高网络运行速度;其次,引入CA(Coordinate Attention)坐标注意力机制,补充位置信息,增强关键信息的特征表达;最后,使用DIoU损失函数替换原YOLOv5s中的GIoU损失函数,提升算法的目标检测能力。结果表明,在自建骑行电动车头盔数据集上,改进后的YOLOv5s网络对骑行头盔的检测平均精度比原始YOLOv5s提高了3.7%,能够实现对骑行头盔佩戴的检测。

E-bikes have gradually become a mainstream mode of transport due to their convenience.However,the casualty rate of e-bike riders in traffic accidents remains high.To solve the problem of the high casualty rate of e-bikes,the modified YOLOv5s is used to detect the wearing of helmets in road scenarios.Firstly,the GSConv module is introduced to replace the standard convolution in the original YOLOv5s backbone network,which improves the network speed while guaranteeing detection accuracy.Secondly,the CA(Coordinate Attention)mechanism is introduced to supplement the position information and improve the feature expression of the key information.Finally,the DIoU loss function is used to replace the GIoU loss function in the original YOLOv5s,improving the ability of target detection by the algorithm.The experimental results show that on the self-constructed dataset of e-bike riding helmets,the average accuracy of e-bike riding helmet detection by the modified YOLOv5s network improves by 3.7%compared to the original YOLOv5s,which indicates that the modified YOLOv5s network can realize the task of detecting the wearing of cycling helmets.