Research on Traffic Sign Recognition Method Based on Generating Countermeasures Network

Abstract

Nowadays, intelligent assisted driving technology is developing rapidly. In the field of intelligent assisted driving technology, traffic sign recognition is a crucial area which focuses on identifying input images and outputting corresponding traffic sign types in the images. However, due to the limitations of optical hardware in the image acquisition system or the poor environment in which the image acquisition system operates, the collected images may not be well recognized by classifiers used for traffic sign classification, despite containing traffic sign information. This could be due to low resolution of images, unclear images, and excessive noise. This article presents a modified generative adversarial network for reconstructing high-resolution traffic signs, specifically suited for the reconstruction of traffic sign images. The generated images and interpolated images are then sent into a convolutional neural network, which is a classifier, to complete the task of classifying traffic signs. The accuracy of the two sets of data is compared, and it is concluded that the classification accuracy of images reconstructed by the generative adversarial network is generally 15% higher than that of images processed by the interpolation method. This study not only proves that the low resolution of images containing traffic sign information can negatively affect subsequent traffic sign classification tasks, but also demonstrates that traffic sign recognition methods based on generating adversarial networks can indeed improve classification accuracy.