Pavement Marking Worn Identification and Classification Using Low-Channel LiDAR

Abstract

Pavement marking retroreflectivity and diffuse illumination can degrade due to wear, cracks, and aging. To enable efficient, safe, and cost-effective inspections of pavement markings, and to ensure timely maintenance, regulate driver behavior, and enhance traffic safety, a method using an onboard low-channel light detection and range (LiDAR) for detecting and classifying worn pavement marking was proposed. The process begins by applying coordinate transform, ground mapping, and sigmoid function filtering to the collected point cloud data to differentiate pavement markings from asphalt pavement. The sparse point cloud is then divided into grids, formatted into a matrix, and missing values are filled in to generate a grayscale map of the pavement marking matrix. Worn areas are segmented using the OTSU and seed region growing (SRG) algorithms and classified into four categories: penetrating, invasive, edge, and internal disease. Field tests showed that the method achieved average worn detection precision, recall, and ${F}1$ -score of 0.8372, 0.8412, and 0.8389, respectively.