Instance Segmentation of Road Marking Signs Using YOLO Models

-Recently, Taiwan has witnessed a significant rise in the number of vehicles, including cars and motorcycles, leading to increased traffic accidents. In many instances, unclear or improperly marked road markings have led drivers to misjudge driving directions, resulting in accidents and penalties. Addressing the challenge, our study focuses on developing a system for detecting road markings, which can help build an Advanced Driving Assistant System (ADAS) and reduce the number of accidents caused by drivers' negligence of road marking signs. We employed and compared the performance of YOLOv5n-seg and YOLOv8nseg, two versions of You Only Look Once (YOLO) version for instance segmentation. We also compiled and proposed our dataset for instance segmentation of Taiwan road marking signs. Our research shows that YOLOv8n-seg performs better than YOLOv5n-seg in segmenting Taiwan road marking signs. YOLOv8n-seg also converges faster during training, leading to shorter training time than YOLOv5n-seg.