摘要
道路交通标志在指导安全出行方面起了重要作用,随着智能交通的发展,交通标志识别越来越受到重视。不同光照、雾气下的复杂自然场景(如树林、建筑物)对交通标志识别干扰较大,为减少这些无关干扰因素所带来的识别率不高的问题,提出了一种语义分割网络与分类网络级联的交通标志识别方法。语义分割网络由UNet改进得到,利用了交通标志与背景颜色、形状特征的差异实现对交通图像感兴趣区域的准确提取;而分类网络则是借鉴LeNet5设计的网络结构,在交通标志感兴趣区域的基础上进行特征提取并分类。实验过程中选取三角形和圆形标志构建数据集,实验结果表明,文中方法与其他较好的交通标志分类方法如HOG-SVM、ResNet50相比,在识别时间较短的同时,其识别精度达到了98.96%。
The traffic signs on road play an important role in guiding safe travel.With the development of intelligent transportation,traffic sign recognition has been paid more and more attention.The complex natural scenes(such as trees and buildings)under different illumination and fog have great interference on traffic sign recognition,in order to reduce the problem of low recognition rate caused by these irrelevant interference factors,a method that traffic sign recognition is realized by cascading semantic segmentation network and classification recognition network is proposed.The semantic segmentation network is obtained by improving UNet,which uses the differences of color and shape features between traffic signs and background to extract the region of interest accurately;while the classification network uses the network structure of lenet5 for reference,which extracts features and classifies traffic signs on the basis of ROI.During the experiment,triangle and circle traffic signs were selected to construct the data set,it showed that the recognition time of this method is shorter than other better traffic sign classification methods HOG⁃SVM and ResNet50,the recognition accuracy achieved 98.96%at the mean time.
作者
张博
ZHANG Bo(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510640,China)
出处
《电子设计工程》
2022年第3期20-25,30,共7页
Electronic Design Engineering
基金
广东省重点领域研发计划项目(2019B090912001)。
关键词
交通工程
卷积神经网络
图像语义分割
感兴趣区域
交通标志识别
traffic engineering
convolution neural network
image semantic segmentation
region of interest
traffic sign recognition
