Predicting Damage of Dutch Road Markings

Amanda Jansen, Vivienne Jansen, Jurjen Helmus, Arnoud Visser

Track: Type A (Regular Papers)

Keywords: computer vision, segmentation, classification

Abstract: Road markings play a crucial role in road safety by guiding traffic and ensuring visibility. As markings deteriorate over time, their effectiveness diminishes, necessitating timely maintenance. This paper studies two methods to classify road-marking damage from recorded images, in accordance with the Dutch CROW guidelines. The first is a model based approach, which first uses a regression model to estimate the marking damage, and then applies the thresholds in the CROW guidelines to classify the damage class. In contrast, a data-driven approach is used, classifying directly the damage class with a YOLOv8 classifier. The data-driven approach achieves an F1-score of 0.97 for the binary-classification task and 0.75 for the multiclass classification task. Compared to other international studies, this is a competitive result.

Serve As Reviewer: ~Arnoud_Visser1

Submission Number: 28