基于改进YOLOv5的交通标志检测与识别  

Traffic Sign Detection Algorithm Based on Improved YOLOv5

作　　者：刘昕宇 薛波 林梦成 Liu Xinyu;Xue Bo;Lin Mengcheng(School of Mechanical Engineering,Jiangsu Institute of Technology,Changzhou,213001,China)

机构地区：[1]江苏理工学院机械工程学院,江苏常州213001

出　　处：《台州学院学报》2024年第6期47-55,共9页Journal of Taizhou University

基　　金：国家自然科学基金项目(62003151);江苏省基础研究计划项目(BK20191035)。

摘　　要：针对无人驾驶时识别不同道路环境下小型交通标志准确率低、检测速度慢的问题,提出一种基于改进YOLOv5的目标识别算法。为了减小模型体积,提高模型推理速度,该算法用Ghost模块替换原有的网络架构,在特征融合阶段结合通道注意力机制,以帮助模型更好地聚焦于图像中的关键信息。为了增强对小目标的特征提取能力,对原C3模块进行改进,引入滑动窗口模块(Swin Transformer Block,STB),并采用中国交通标志数据集TT100K进行对比验证。结果表明:文中提出的方法识别准确率为92.6%,比原始YOLOv5算法提升了0.7%;平均准确率均值达到93.5%,比YOLOv5算法提高了2.5%。Aiming at the problems of low accuracy and slow detection speed in identifying small traffic signs on roads in different environments during driverless driving,a target recognition algorithm based on improved YOLOv5 was proposed.In order to reduce the model volume and improve the model inference speed,this algorithm uses the Ghost feature module to replace the original network architecture.Fusion and channel attention mechanisms are combined at the feature module level to help the model better focus on key information in the image.In order to enhance the feature extraction ability of small targets,the original C3 module was improved,the STB module(Swin Transformer Blocks)was introduced,and the Chinese traffic sign data set TT100K was introduced for comparison and verification.The results showed that the recognition accuracy of the method proposed in the paper was 92.6%,an increase of 0.7%;The value reached 93.5%,an increase of 2.5%.

关 键 词：交通标志识别 目标检测 注意力机制 

分 类 号：U463.6[机械工程—车辆工程] TP391.41[交通运输工程—载运工具运用工程]