基于YOLOv5s的自动驾驶车辆行人检测方法
Pedestrian Detection Method for Autonomous Driving Vehicles Based on YOLOv5s
侯佩玉 1徐淼 1张明 1柳庆 1徐芳冬1
作者信息
- 1. 北华大学土木与交通学院,吉林 吉林 132013
- 折叠
摘要
提出一种改进的YOLOv5s模型,旨在提高自动驾驶车辆行人检测的准确性.通过增加一个小目标检测层,将原有的三尺度检测层扩展为四尺度,提高小目标的检测能力;在颈部网络中将坐标注意力(CA)机制引入C3 模块,构建C3-CA模块,以强化特征间的空间关系,从而更精确地定位行人;将原有的CIoU损失函数替换为EIoU,改善模型收敛性.在BDD100K数据集进行试验验证,结果显示:与YOLOv5s模型相比,改进后的模型检测精度提高了1.7%,召回率提高了4.8%,平均精度提高了2.2%,降低了漏检和误检概率.
Abstract
An enhanced YOLOv5s model is proposed to improve pedestrian detection accuracy in autonomous driving applications.This enhancement includes expanding the original three-scale detection head to a four-scale detection structure by adding an additional layer specialized in small object detection,thereby augmenting the model's capability to detect small-scale objects.In the neck network,a Coordinate Attention(CA)mechanism is integrated into the C3 module,forming a C3-CA module to reinforce spatial dependencies between features,which enables more precise pedestrian localization.Furthermore,the original CIoU loss function is replaced with an EIoU loss function to enhance model convergence.Experiments conducted on the BDD100K dataset demonstrate that the improved model,compared with the baseline YOLOv5s,achieves 1.7%increase in detection precision,4.8%boost in recall,and 2.2%improvement in mean average precision,effectively reducing missed and false detections.
关键词
自动驾驶/行人检测/YOLOv5s/坐标注意力Key words
autonomous driving/pedestrian detection/YOLOv5s/coordinate attention引用本文复制引用
出版年
2025
国家科技期刊平台
NSTL
万方数据