Hough-RANSAC-YOLO: A Hybrid Approach for Robust Traffic Light Detection in Real-World Conditions

Abstract

Accurate and robust detection of traffic lights in real-world conditions remains a major challenge for computer vision systems, particularly in the context of autonomous vehicles and advanced driver assistance systems. This paper introduces HoughRANSAC-YOLO, an innovative hybrid approach that combines the robustness of traditional computer vision techniques such as the Hough Transform and RANSAC with the power of deep learning offered by YOLOv8. Our method begins by employing the Hough Transform to detect circular shapes, followed by RANSAC to effectively eliminate false positives. Finally, it leverages the speed and accuracy of YOLOv8 for object detection. Thorough experiments on diverse datasets, including complex urban scenarios and various weather conditions, show that HoughRANSAC-YOLO surpasses existing methods in terms of accuracy and resilience to partial occlusions and changes in lighting. Our approach significantly reduces the rate of false positives while maintaining processing times suitable for real-time applications. These findings suggest that a judicious integration of classical techniques with deep learning can substantially enhance the reliability of traffic light detection systems, thereby improving road safety and advancing autonomous driving technologies.