European Journal of Molecular & Clinical Medicine

Abstract

Due to the large number of deaths and car accidents caused by the driver's lack of attention, car manufacturers are trying to integrate ADAS systems with artificial intelligence and CV. One function that helps the driver is traffic sign recognition (TSR). This is a technology with which a vehicle is able to recognize road signs placed on the road, e.g. "speed limit" or "give way" or "stop" all this being possible with the help of computer vision and Convolutional Neural Networks. In this article we propose an implementation based on LeNet architecture for traffic sign recognition using CNN. We preferred the deep learning approach for this challenge as methods like shape based or color based share a common weakness in factors as light changes, scale change, rotation. The paper presents implementation of the network architecture based on LeNet5 using Keras and TensorFlow library, how we trained the CNN using the GTSRB dataset, and what results we obtained training and using the network for real time applications using only CPU power, with Intel i7 7700K. The experimental results showed a 95% accuracy in recognizing traffic signs. Keywords – Artificial Intelligence, AI, Convolutional neural network, deep learning, LeNet architecture, traffic signs recognition.