Point-Based Visual Status Evaluation of Worn Pavement Markings Based on a Feature-Binary-PointNet Network and Shape Descriptors Using LiDAR Point Clouds: A Case Study of an Expressway

Abstract

Inspecting the visual wear status of pavement markings is important to guide the maintenance plans of roads and ensure traffic safety. This study proposes a point-based visual status evaluation system for pavement markings using light detection and ranging (LiDAR) technology. The system includes a feature-binary-PointNet (FBP) network to accurately extract pavement markings from raw point clouds. Furthermore, multi-shape descriptors are constructed to efficiently identify different types of pavement markings, and a four-level wear-grading system is established based on two indices (point-based visual wear area and wear area proportion) of each pavement marking. The proposed method was tested using the mobile laser scanning technique considering a case study of an expressway. The proposed FBP network performed better than the existing state-of-the-art networks in extracting the pavement markings, with accuracy, recall, precision, and mean intersection over union scores of 97.04, 84.42, 91.11, and 79.36%, respectively. The main types of expressway markings were rapidly identified, and maintenance advice was provided based on the two computed indices and the wear grade of each marking. The study findings can guide managing agencies in allocating the maintenance budget for fixing highly defective road sections.