Machine detectability of road markings using classical image processing techniques for demand-oriented road operations in automated vehicles

This paper investigates the Machine Detectability (MD) of road markings under various environmental conditions, crucial for the definition of Operational Design Domains (ODD) of automated driving systems as well as the assessment of Operational Domains (OD). By analysing the correlation between MD parameters, specifically contrast, gradient, and edge detectability, and common photometric properties of road markings currently used for maintenance management (retroreflectivity (R_L) and daytime visibility (Q_d)), the paper aims to bridge the gap in current road marking detectability research and OD assessment. The methodology encompassed a detailed examination of road markings on a motorway under different lighting and weather conditions, employing both camera and LiDAR sensors for data collection. The findings reveal that R_L is a consistent predictor for MD in camera images during nighttime and for LiDAR contrast in dry and moist conditions, whereas Q_d fails to reliably predict MD in daytime conditions. Moreover, the study introduces a multi-parameter approach that transcends sole contrast analysis as well as the usage of off-the-shelf machine vision systems, proposing a new set of MD parameters for a broader and transparent evaluation of road marking detectability. This comprehensive assessment highlights the need for quality standards for road markings that would accommodate varying environmental impacts on MD of road markings. Ultimately, this research provides valuable insights and recommendations on research approaches to find demand-oriented minimum standards for MD of road markings, enabling comprehensive OD assessments, and facilitating safer navigation for automated vehicles.