Traffic Sign Classification Using Deep Learning Based Convolutional Neural Networks

Abstract

Traffic sign classification is a challenging computer vision task of high industrial relevance. Traffic signs contain important road traffic information and are considered the cornerstone of traffic systems as they ensure road safety for both pedestrians and drivers. Traffic sign classification is considered a crucial part in the operation of advanced driver assistance systems, autonomous vehicles, and intelligent transportation systems. Traffic sign classification helps reduce the number of road accidents and wrong decisions made by drivers and helps foster the credibility of autonomous vehicles. In this paper, a computer-aided recognition system was designed for the purpose of traffic sign classification.The system used deep learning, specifically, convolutional neural networks. The proposed model was based on a reduced version of the VGG16 model while integrating batch normalization, dropout and Adam optimization, Google Colaboratory was used during training as it offers the use of virtual machines with GPUs that help speed up the training process immensely. Simulation results show that the proposed model performed in an excellent manner, reaching evaluation metric values of above 97%.