A framework for road space extraction from point clouds and integration into 3D city models

Abstract

This study presents a robust methodology for segmenting an urban road network at multiple levels of granularity, leveraging Mobile Laser Scanning (MLS) point cloud data and the CityGML 3.0 framework. The proposed approach integrates semantic and geometric information to delineate road spaces into sections, intersections, sidewalks, parking areas, and individual driving lanes. The methodology achieves spatial accuracy and compliance with road design standards through the use of clustering algorithms, alpha shape methods, and geometric refinements. Case studies in two 2 km intra-urban networks—Santiago de Compostela and Madrid, both comprising local and collector streets with uninterrupted carriageways and white-colour markings—demonstrate the approach’s effectiveness, yielding highly accurate results with average Intersection over Union (IoU) scores of 0.85 for Santiago de Compostela and 0.83 for Madrid across road features. Parking areas in Santiago de Compostela achieved 0.9 IoU, while zebra crossings in Madrid exhibited limitations with 0.68 IoU due to their smaller size and complex geometry. The results provide a standardized and detailed road network representation, suitable for urban planning, traffic management, and autonomous vehicle navigation. Future work aims to expand the methodology for diverse datasets, incorporate multi-temporal analyses, and integrate additional traffic objects for enhanced CityGML modelling. This research highlights the potential of point cloud-based methods to advance digital twin development and 3D semantic urban modelling.