Traffic Sign Recognition with a small convolutional neural network

Abstract

Abstract Traffic sign recognition is an important part of the intelligent transportation system and has important application prospects in driverless vehicles and driver assistance systems[1]. In the image recognition of traffic signs, according to the image features of traffic signs, the common methods include traditional template matching method[2], SVM method[3], random forest[4] and the best Convolutional Neural Networks (CNN) method. In this paper, a new CNN is proposed. Feature extraction, compared with the traditional CNN method, has higher accuracy, fewer parameters, smaller models and easier training, which is evaluated on the German Traffic Sign Recognition Benchmark (GTSRB) and the Belgium Traffic Sign Dataset (BTSD). The results show that this method is superior to the traditional CNN method in traffic identification.