Traffic Sign Recognition via Transfer Learning Using Convolutional Neural Network Models

Abstract

Traffic sign recognition is one of the most important applications for advanced driving support systems. Studies on deep learning in recent years have increased considerably in this area. Although high accuracy is achieved with deep learning, it requires a lot of data sets, training of these data sets takes a lot of time and turns into a laborious task. However, a considerable advantage in terms of time and performance can be achieved by using pre-trained models with the transfer learning method. In this study, some improvement processes were performed on pre-trained convolutional neural network models with ImageNet database. Then, the recognition process was performed for 10 classes in the GTSRB database. The models used here are VGG19, ResNet, MobileNet and Xception. When the results are compared, it is seen that the best accuracy value is achieved with MobileNet model.