An Enhanced YOLO Architecture for Robust Real-Time Traffic Sign Detection

Traffic sign detection plays a vital role in enhancing road safety and enabling autonomous driving systems. This project presents an enhanced version of the YOLO (You Only Look Once) architecture, specifically YOLOv12, designed for robust real-time traffic sign detection. The proposed model is optimized for improved accuracy and speed, with the ability to detect a wide variety of traffic signs including hazard signs, roundabouts, speed limits, stop signs, and traffic lights, under various environmental conditions. The YOLOv12 model leverages state-of-the-art convolutional layers and attention mechanisms to capture fine-grained details of traffic signs from live video streams. The model is evaluated on a custom dataset containing 402 images and 487 instances, with impressive results: a mean Average Precision (mAP) of 0.94 at IoU threshold 0.50, and a mAP50-95 of 0.668 across multiple classes. Specific traffic sign classes, such as hazard signs and speed limits, achieved near-perfect detection accuracy, with precision and recall values reaching 1.0 for several categories. These results highlight the model’s capability to effectively identify traffic signs in challenging conditions. This work provides a foundation for integrating enhanced YOLO models in real-time traffic monitoring and autonomous vehicle systems. By combining deep learning techniques with cutting-edge computer vision, the project demonstrates the potential of AI-driven solutions in transforming road safety applications.

Keywords: YOLOv12, Traffic Sign Detection, Deep Learning, Computer Vision, Autonomous Driving, Real-Time Detection, Object Detection, Mean Average Precision (mAP), Traffic Safety, Machine Learning, Precision and Recall, AI in Traffic Monitoring.