Automatic Extraction of Road Curb and Road Surface from Lidar Point Cloud Data

Abstract

Road feature extraction is crucial for wide range of geospatial applications such as road management, intelligent transportation and road safety evaluation. Due to the efficient vehicle-based on-road scanning opportunity, Mobile Laser Scanning (MLS) has become the most appropriate data acquisition system for road environments. Most of methods available for road feature extraction are classical approaches that do not mitigate the problems caused by the presence of outliers and occlusions. This study proposed an automated method for extraction and vectorization of road surface and curb, utilizing a grid-based segmentation and classification. The method begins with extracting the ground surface from the point cloud, where an elevation-based thresholding and Cloth Simulation Filtering (CSF) are applied to isolate terrain points. A region growing based segmentation algorithm is applied to identify the road surface and curb structures based on the elevation differences and surface normal orientations. To address main challenges in region growing, such as curb cuts and occlusions from parked vehicles, a local re-segmentation is developed. The local edge detection and gap bridging are applied for efficient region-growing. The extracted curb boundaries are then vectorized using the alpha-shape algorithms, ensuring a structured, GIS compatible representation. The developed algorithm was tested on multiple urban datasets, and reached a classification accuracy of 98% from the constructed confusion matrix. The geometric accuracy exceeded 85% for most curb islands, with some achieving 97.7% precision. The results validated the robustness and scalability of the developed method in urban environments, and provided a computationally efficient solution for automated road and curb extraction.