This paper presents a comparative analysis of YOLOv8 and RT-DETR object detection models within the context of Advanced Driver Assistance Systems (ADAS). The study evaluates both models across five datasets, focusing on mean Average Precision (mAP), inference speed (Frames Per Second, FPS), and F1 scores. YOLOv8 models, particularly YOLOv8x, demonstrated superior performance, achieving faster convergence, higher detection performance, and lower latency compared to RT-DETR, making them more suitable for real-time applications. The analysis also highlights YOLOv8’s effectiveness in handling class imbalance, with higher F1 scores reflecting a better balance between precision and recall. RT-DETR models, while showing potential, lagged in both detection performance and inference time. The findings suggest that YOLOv8 is the better choice for real-time ADAS tasks, offering a strong balance between detection performance and processing efficiency.