Exploration of Traffic Sign Recognition Using AI and Computer Vision

Abstract

Traffic Sign Recognition (TSR) is a critical perception module in modern Advanced Driver Assistance Systems (ADAS) and autonomous vehicle technology. Ensuring road safety and vehicle autonomy depends on the accurate and real- time detection and classification of traffic signs. This paper presents a novel, cost-effective system for robust traffic sign recognition using an ESP32-CAM module for image acquisition and the YOLOv3 algorithm for object detection. A key focus of this project is to address real-world challenges where signs may be partially obscured by environmental factors like mud or water. The proposed system captures an image via the ESP32-CAM, transmits it through a Telegram bot for easy access, processes the image to enhance clarity, and then uses a trained YOLOv3 model to identify the traffic sign. The results are displayed on a user-friendly GUI, which also facilitates remote control of a vehicle prototype equipped with an L298N motor driver. The model, trained on the German Traffic Sign Recognition Benchmark (GTSRB), achieves high accuracy and demonstrates the viability of using integrated IoT and deep learning to enhance road safety under challenging conditions.