Customized convolutional neural networks for Moroccan traffic signs classification

Abstract

Recognition of traffic signs is a challenging task that can enhance road safety. Deep neural networks have demonstrated remarkable results in numerous applications, such as traffic signs recognition. In this paper, we propose an innovative and efficient system for recognizing traffic signs, based on customized convolutional neural network (CNN) developed through hyperparameters optimization. The effectiveness of the proposed system is assessed using a novel dataset, the Moroccan traffic signs dataset. The results show that the proposed design recognizes traffic signs with an accuracy of 0.9898, outperforming several CNN architectures such as VGGNet, DensNet, and ResNet.