摘要
在自动驾驶领域准确实时地检测出小目标交通标志具有重要意义,本文针对YOLOv5s算法检测小目标交通标志精度低、漏检等问题,提出了一种基于改进YOLOv5s的交通标志检测算法。将Transformer编码结构与C3模块结合,用新的C3TR替换主干网络中最后一个C3模块,提高主干网络对图像全局特征信息的提取能力;用EIoU Loss替换YOLOv5s的定位损失函数,提高模型检测框的回归精度;在多尺度检测部分,通过增加一层浅层检测层作为更小目标的检测层,提高对小目标交通标志的检测能力。实验结果表明,改进YOLOv5s检测算法在CCTSDB数据集上的平均检测精度(mAP)为93.1%,比原YOLOv5s提升了3.6%,对小目标交通标志检测精度更高。
It is of great significance to accurately detect small target traffic signs at real time in the field of autonomous driving.Aiming at problems such as low accuracy and missing detection of small target traffic signs by YOLOv5s algorithm,a traffic sign detection algorithm based on improved YOLOv5s was proposed.Transformer coding structure is combined with C3 module to replace the last C3 module in the trunk network with a new C3TR to improve the trunk network′s ability to extract global feature information of images.EIoULoss was used to replace the positioning loss function of YOLOv5s to improve the regression accuracy of the model detection frame.In the multi-scale detection part,a shallow detection layer is added as the detection layer of smaller targets to improve the detection ability of traffic signs.The experimental results show that the mean precision(mAP)of the improved YOLOv5s detection algorithm on the CCTSDB data set is 93.1%,which is 3.6%higher than the original YOLOv5s detection algorithm,and the detection accuracy of small-target traffic signs is higher.
作者
周晋伟
王建平
阜远远
张太盛
方祥建
王嘉鑫
王天阳
ZHOU Jinwei;WANG Jianping;FU Yuanyuan;ZHANG Taisheng;FANG Xiangjian;WANG Jiaxin;WANG Tianyang(School of Mechanical Engineering,Anhui Polytechnic University,Wuhu 241000,China;Engineering Department,CRRC Puzhen Alstom Transportation Systems Co.,Ltd.,Wuhu 241000,China)
出处
《安徽工程大学学报》
CAS
2024年第2期40-46,共7页
Journal of Anhui Polytechnic University
基金
安徽省科技重大专项项目(202103a05020033)。
作者简介
周晋伟(1997-),男,安徽阜阳人,硕士研究生;通信作者:王建平(1970-),男,甘肃天水人,教授,硕导。
