ABSTRACT
The main objective of this work is to propose an efficient approach applicable to Intelligent Transportation Systems (ITS), particularly embedded vision systems, capable of detecting and tracking objects in a dynamic environment. Many previous works have been conducted in this field to design a powerful and robust ITS system taking into account the notion of optimization. However, in this work, we aim to develop an intelligent system based on Computer Vision (CV) and Artificial Intelligence (AI) that combines two distinct phases respectively; namely detection and recognition. To achieve this objective, the paper focuses on the detection and recognition of traffic road signs. The proposed approach is based on an algorithm named Easy Road Sign Detection and Recognition Algorithm (ERSDRA). To detect traffic signs, we relied on colors (red, blue) and shapes (circle, square, triangle, and rectangle), and for traffic sign recognition, we proposed a CNN architecture (Convolution Neural network). The design of this architecture was inspired by two fundamental architectures, namely LeNet-5 and AlexNet, in order to have an optimistic CNN model after Deep Learning (DL). This CNN model is employed for the recognition of each cropped traffic sign, while Grad-CAM is used as explainable AI techniques to interpret visually the results of the CNN model. Finally, after several tests carried out on our improved approach, for the detection module, the error rate range is between 0,001% and 0,025% for triangular and square shapes respectively, while the recognition module reached the best score of 98,47% in accuracy. Given the performance and interpretability offered by the proposed model, we can say that our model can be useful in the automotive field for road safety, especially in Advanced Driving Assistance Systems (ADAS).
