Abstract
Developments in three-dimensional real worlds promote the integration of geoinformation and building information models (BIM) known as GeoBIM in urban construction. Light detection and ranging (LiDAR) integrated with global navigation satellite systems can provide geo-referenced spatial information. However, constructing detailed urban GeoBIM poses challenges in terms of LiDAR data quality. BIM models designed from software are rich in geometrical information but often lack accurate geo-referenced locations. In this paper, we propose a complementary strategy that integrates LiDAR point clouds with as-designed BIM models for reconstructing urban scenes. A state-of-the-art deep learning framework and graph theory are first combined for LiDAR point cloud segmentation. A coarse-to-fine matching program is then developed to integrate object point clouds with corresponding BIM models. Results show the overall segmentation accuracy of LiDAR datasets reaches up to 90%, and average positioning accuracies of BIM models are 0.023 m for pole-like objects and 0.156 m for buildings, demonstrating the effectiveness of the method in segmentation and matching processes. This work offers a practical solution for rapid and accurate urban GeoBIM construction.
Introduction
In urban planning and "Digital Twin" construction, a significant data mismatch exists between Building Information Models (BIM) and Geographic Information Systems (GIS). BIM models are rich in geometric and semantic detail but typically lack accurate geo-referenced coordinates. Conversely, LiDAR point clouds provide excellent spatial data but are often noisy and unstructured. Integrating these into a "GeoBIM" environment usually requires tedious manual alignment. This research presents an automated pipeline to bridge this gap, leveraging weakly supervised deep learning and a coarse-to-fine matching strategy to anchor high-detail architectural models into accurate real-world coordinates.
Methodology
The proposed framework utilizes a two-stage complementary strategy to automate the fusion of LiDAR data and BIM models.
1. Semantic and Instance Segmentation
The first stage involves segmenting the raw LiDAR point cloud into meaningful objects. To avoid the labor-intensive process of manual labeling, the authors employ geometric heuristics—such as identifying poles as vertical lines and buildings as smooth planes—to automatically generate training samples.
A weakly supervised learning approach is then applied using a Kernel Point Convolution (KPConv) backbone. This network is trained on the automated labels to perform semantic segmentation. To refine these results and move from point-level classification to object-level instances, graph-based clustering is utilized. This technique clusters points based on structural properties like distance and anisotropy, effectively separating adjacent objects like trees and buildings.
2. Coarse-to-Fine Matching
Once objects are segmented, they are matched with their corresponding "as-designed" BIM models. For buildings, a modified 4-Points Congruent Sets (4PCS) algorithm provides coarse alignment, while pole-like objects are aligned via Principal Component Analysis (PCA). Fine registration is then achieved using a point-to-plane Iterative Closest Point (ICP) variant, which ensures the BIM models "snap" into the correct geographic location.
Results and Discussion
The framework was validated using two datasets: a backpack LiDAR scan from Hong Kong and the Paris-Lille-3D benchmark. The system achieved an overall segmentation accuracy of over 90%.
Crucially, the positioning precision reached centimeter-level accuracy:
- Pole-like objects: Average error of 0.023 m.
- Buildings: Average error of 0.156 m.
The "complementary" philosophy proved robust even with incomplete LiDAR data (e.g., missing roofs), as the matching algorithm effectively utilized available facade information to constrain the BIM models.
Conclusion
This work provides a practical solution for rapid and accurate urban GeoBIM construction. By using LiDAR as a "spatial anchor" for existing high-quality models, the method bypasses the need for expensive reconstruction from scratch. Future work will focus on scaling this approach to city-wide databases and improving the handling of complex vegetation and architectural geometries.
Summary updated