YOLOv8m: A Deep Learning Approach to Traffic Sign Recognition in CARLA and Speed control of Autonomous Vehicles

Traffic sign detection acts as the eyes for autonomous vehicles, employing computer vision to decipher road signs for safe and compliant navigation. This technology tackles challenges like variable lighting and occlusions by leveraging deep learning models trained on diverse datasets. By interpreting signs accurately, autonomous vehicles can navigate roads confidently, paving the way for a safer future. Traffic sign detection in real-world scenarios faces a confluence of challenges: varying lighting and weather conditions can degrade sign visibility, partial or complete occlusions by other objects can hinder detection, and the sheer diversity of traffic signs across regions necessitates robust models capable of generalizability. The research utilizes a combined real-world (65%) and simulated (CARLA, 35%) dataset for training a YOLOv8m model for speed limit sign detection (30, 60, 90 km/h) in autonomous vehicles. Image augmentation and collaborative annotation via Makesense platform enrich the dataset. The model has achieved 98% accuracy rate in detecting speed limits on the road side, even in difficult situations. The model is evaluated in the CARLA simulator for controlled testing before real-world implementation