Road Marking Detection and Instance Segmentation Using YOLOv8 Models

Abstract

This paper investigates the concepts of road marking detection and instance segmentation in the context of autonomous driving. It discusses the significance of these tasks and their applications in traffic analysis and road safety. The paper highlights the role of instance segmentation in enhancing object detection, tracking, semantic mapping, and scene understanding in autonomous driving systems. It emphasizes the benefits of instance segmentation in improving precision, handling occlusions, and enhancing association and track maintenance. The paper also evaluates the YOLOv8 model's performance in road marking detection and instance segmentation. The evaluation dataset comprises 2885 images with 11 classes of road markings captured in diverse weather conditions. Various sizes of YOLOv8 models are thoroughly assessed to measure their effectiveness in addressing these tasks.