Detection of Red, Yellow, and Green Lights in Real-Time Traffic Lights with YOLO Architecture

Authors : Abdil Karakan

Pages : 28-36

Doi:10.18466/cbayarfbe.1432356

View : 86 | Download : 106

Publication Date : 2024-06-28

Article Type : Research Paper

Abstract :In the study, red, yellow, and green lights at traffic lights were detected in real-world conditions and in real time. To adapt to real-world conditions, A data set was prepared from traffic lights in different locations, lighting conditions, and angles. A total of 5273 photographs of different traffic lights and different burning lamps were used in the data set. Additionally, grayscale, bevel, blur, variability, added noise, changed image brightness, changed color vibrancy, changed perspective, and resized and changed position have been added to photos. With these additions, the error that may occur due to any distortion from the camera is minimized. Four different YOLO architectures were used to achieve the highest accuracy rate on the dataset. As a result, the study obtained the highest accuracy at 98.3% in the YOLOV8 architecture, with an F1-Score of 0.939 and [email protected] value of 0.977. Since the work will be done in real time, the number of frames per second (FPS) must be the highest. The highest FPS number was 60 in the YOLOv8 architecture.
Keywords : Deep learning, traffic light, real time detection, YOLOV8