基于Transformer改进YOLOv5的交通标志检测算法  

Improved traffic sign detection algorithm based on Transformer and YOLOv5

作　　者：韩长江 刘丽娟[1] HAN Chang-jiang;LIU Li-juan(Software Technology Institute,Dalian Jiaotong University,Dalian 116052,Liaoning Province,China)

机构地区：[1]大连交通大学软件学院,辽宁大连116052

出　　处：《信息技术》2024年第11期21-27,共7页Information Technology

基　　金：辽宁省自然科学基金面上项目(2022-MS-341)。

摘　　要：交通标志检测作为自动驾驶的组成部分直接影响着行车安全。针对现有算法对图像中尺寸小、被遮挡的标志存在漏检、误检的问题,文中提出了基于改进YOLOv5的交通标志检测算法。首先对原模型注意力缺失的问题经过对比后构建了BiFormer-y,使模型可以更好获取长期依赖;接着针对层数较深造成的具有丢失特征的缺陷,利用残差结构重新设计检测层,从而更好地保留特征;最后对耦合头的空间错位问题引入解耦头并进行优化。CCTSDB2021的实验表明,精确率、召回率、mAP分别为97.0、95.9、97.9与先进工作相比具有明显优势。Traffic sign detection,as an integral part of autonomous driving,directly affects traffic safety.In this paper,a traffic sign detection algorithm based on the improved YOLOv5 is proposed to address the issues of missed detection and false detection of small-sized signs in images.Firstly,the problem of attention deficiency in the original model is addressed by comparing and constructing BiFormer-y,which allows the model to better capture long-term dependencies.Then,to address the issue of feature loss in deeper layers,a residual structure is utilized to redesign the detection layers,thereby preserving features more effectively.Finally,to tackle the problem of spatial misalignment in coupled heads,a decoupled head is introduced and optimized.Experiment results on CCTSDB2021 demonstrate significant advantages over state-of-the-art methods,with precision,recall,and mAP reaching 97.0,95.9,and 97.9,respectively.

关 键 词：机器视觉 目标检测 TRANSFORMER YOLOv5s算法 交通标志 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]