基于YOLOv5s算法在行人车辆图像检测方面的改进和研究  

IMPROVEMENT AND RESEARCH ON PEDESTRIAN AND VEHICLE IMAGE DETECTION BASED ON YOLOV 5 S ALGORITHM

作　　者：朱培瑾 刘平[1] 马学文 ZHU Pei-jin;LIU Ping;MA Xue-wen(School of Computer and Software,Nanyang Institute of Technology,Nanyang 473004,China)

机构地区：[1]南阳理工学院计算机与软件学院,河南南阳473004

出　　处：《南阳理工学院学报》2024年第4期83-89,共7页Journal of Nanyang Institute of Technology

基　　金：河南省科技攻关计划项目(142102210554);南阳市科技攻关计划项目(23KJGG007)。

摘　　要：针对城市交通环境复杂,行人和车辆检测结果存在漏检、误检、精度不高的问题,提出一种基于YOLOv5s的改进模型。改进模型采用EIoU损失函数,并引入了FocalEIoU损失函数,最后通过选用PASCAL VOC数据集来检验改进算法的效果。最终实验结果表明,改进后的YOLOv5s算法mAP提升4%左右,在检测精度上也有所提高,验证了本次研究的有效性。An improved YOLOv5s model is proposed to address the complex urban traffic environment,as well as the issues of missed detection,false detection,and low accuracy in pedestrian and vehicle detection results.The improved model adopts the EIoU loss function and introduces the FocalEIoU loss function.Finally,the effectiveness of the improved algorithm is tested by selecting the PASCAL VOC dataset.The final experimental results show that the improved YOLOv5s algorithm improves mAP by about 4%and also improves detection accuracy,verifying the effectiveness of this study.

关 键 词：YOLOv5s 图像处理 目标检测 行人车辆检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]