Towards a Digital Twin of Liege: The Core 3D Model based on Semantic Segmentation and Automated Modeling of LiDAR Point Clouds

Abstract

The emergence of Digital Twins in city planning and management marks a contemporary trend, elevating the realm of 3D modeling and simulation for cities. In this context, the use of semantic point clouds to generate 3D city models for Digital Twins proves instrumental in addressing this evolving need. This article introduces a processing pipeline for the automatic modeling of buildings, roads, and vegetation based on the semantic segmentation results of 3D LiDAR point clouds. It employs a semantic segmentation approach that integrates multiple training datasets to achieve precise extraction of target objects. Open-source reconstruction tools have been adapted for building and road modeling, while a Python code was optimized for tree modeling, leveraging a foundational code. The case study was conducted in the city of Liège, Belgium. The obtained results were satisfactory, and the schemas and geometry of the developed models were validated. An evaluation of the adopted reconstruction methods was conducted, along with their comparison to other methods from the literature.