Self-driving cars have recently gained in popularity. This is because of rapid advances in vehicle and artificial intelligence technology. Autonomous cars’ ability to drive effectively and safely depends heavily on their capacity to recognize traffic signs. Traditional visual recognition of things, conversely, relies heavily on the extraction of visual features, such as color and edge. Despite these efforts, the varying appearance of road signs across geographical areas, lighting changes, and complex background situations continues to prevent the development of accurate traffic sign recognition platforms. In this paper, we present YOLO-TSR, a novel network based on YOLOv8 that innovatively tackles the challenges encountered in road sign recognition (TSR). Our intention is to provide a method to detect and recognize traffic signs in complex situations and under varying weather conditions. The proposed method was validated against three separate traffic sign datasets: our privately curated dataset, the widely recognized German Traffic Sign Recognition Benchmark (GTSRB) dataset, and the Belgium Traffic Sign Dataset. We conducted numerous experiments to validate the proposed algorithm’s effectiveness. The proposed algorithm achieves 98.79% accuracy, 92.18% recall, 96.21% mAP@0.5, and 84.32% mAP@0.5:0.95 for the GTSRB dataset. For our private dataset, the algorithm had an accuracy of 96.62%, a recall of 90.81%, mAP@0.5 of 94.83%, and mAP@0.5:0.95 of 81.70%. Furthermore, the algorithm maintains a consistent frame rate of 73 frames per second, which meets real-time detection requirements.
YOLO-TSR: A Novel YOLOv8-Based Network for Robust Traffic Sign Recognition
Transportation Research Record: Journal of the Transportation Research Board
2025-04-04
Article (Journal)
Electronic Resource
English
Traffic sign recognition system based on YOLOv8-ConvNeXt
SPIE | 2025
|Traffic Sign Detection Using YOLOv8
IEEE | 2024
|