Traffic sign recognition with object recognition algorithms

The development of autonomous vehicles and intelligent transportation systems has made it essential to detect traffic signs accurately, quickly, and reliably. Real-time recognition of traffic signs is of great importance for vehicles to adapt to road conditions, enhance safety, and comply with traffic regulations. In this thesis, six different versions of YOLOv11, designed to balance both accuracy and computational efficiency, were examined for traffic sign detection. The models were trained on a traffic sign dataset and evaluated using metrics such as precision, recall, and F1 score. Three basic YOLOv11 models and their three improved versions were implemented in the study, aiming to increase detection accuracy. The applied improvements enhanced the generalization capability of the models, allowing them to better detect small or partially occluded signs. This resulted in more reliable performance, especially under complex road conditions and varying lighting environments. The findings of the study demonstrate that the YOLOv11n, YOLOv11s, and YOLOv11m models and their enhanced variants offer viable and powerful solutions for real-time traffic sign detection. The research also highlights that even higher performance can be achieved through data augmentation and fine-tuning techniques, paving the way for future studies. The data augmentation techniques and the carefully conducted labeling process used in this study significantly improved the models' adaptability to real-world conditions. By including data representing different weather and lighting conditions, the robustness of the models increased, and detection errors were minimized. As a result of the study, an F1 score of 0.9882 was obtained for the YOLOv11s model, 0.9877 for the YOLOv11m model, and 0.9899 for the YOLOv11n model. The results indicate that the developed YOLOv11-based models can be used as reliable, fast, and scalable solutions in autonomous driving systems and intelligent transportation applications.