A Lightweight Model for Traffic Sign Classification Based on Enhanced LeNet-5 Network

Abstract

For several years, much research has focused on the importance of traffic sign recognition systems, which have played a very important role in road safety. Researchers have exploited the techniques of machine learning, deep learning, and image processing to carry out their research successfully. The new and recent research on road sign classification and recognition systems is the result of the use of deep learning-based architectures such as the convolutional neural network (CNN) architectures. In this research work, the goal was to achieve a CNN model that is lightweight and easily implemented for an embedded application and with excellent classification accuracy. We choose to work with an improved network LeNet-5 model for the classification of road signs. We trained our model network on the German Traffic Sign Recognition Benchmark (GTSRB) database and also on the Belgian Traffic Sign Data Set (BTSD), and it gave good results compared to other models tested by us and others tested by different researchers. The accuracy was 99.84% on GTSRB and 98.37% on BTSD. The lightness and the reduced number of parameters of our model (0.38 million) based on the enhanced LeNet-5 network pushed us to test our model for an embedded application using a webcam. The results we found are efficient, which emphasize the effectiveness of our method.