Traffic light detection represents a significant issue for Intelligent driving. This problem can be overcome by using the advanced deep learning models, such as the YOLO series that have been extensively used for traffic recognition tasks. This work presents a comparative study of traffic light detection using the YOLO family of object detection algorithms, specially focusing on the latest versions YOLO such as YOLOv8, YOLOv9, and YOLOv10. These versions incorporate significant enhancements in accuracy, model efficiency, and speed, crucial for enhancing Intelligent driving capabilities. The models trained on the cinTA-c2 dataset in Roboflow which includes traffic light images captured in varying conditions (different lighting, weather, and angles). To assess the effectiveness and performance of each YOLO version, we use the object detection metrics such as mean Average Precision (mAP), Recall, and Precision. After training these models on cinTA-c2 dataset, we notice that YOLOv9t is the best model according the obtaining metrics such as Precision, Recall, and mAP.