L2M-Reg: Building-level Uncertainty-aware Registration of Outdoor LiDAR Point Clouds and Semantic 3D City Models

TL;DR

L2M-Reg addresses the challenge of registering outdoor LiDAR point clouds to LoD2 semantic city models at the building level under inherent model uncertainty. It introduces a plane-based framework with three core innovations: automatic representative-region plane extraction, a pseudo-plane-constrained Gauss-Helmert adjustment that decouples 2D horizontal from 3D vertical estimation, and an adaptable vertical translation estimation that leverages ground data when available. By exploiting CityGML semantic information and a Geometric Consistency check, L2M-Reg achieves superior accuracy and efficiency across five real-world datasets, outperforming leading ICP-based and plane-based baselines. The approach enables uncertainty-aware, building-level model updating and demonstrates robust generalizability, while acknowledging limitations related to highly non-planar facades and partial occlusions. Overall, L2M-Reg advances practical, uncertainty-aware LiDAR-to-Model registration and expands the reliable use of LoD2 models in precision-oriented urban tasks.

Abstract

Accurate registration between LiDAR (Light Detection and Ranging) point clouds and semantic 3D city models is a fundamental topic in urban digital twinning and a prerequisite for downstream tasks, such as digital construction, change detection, and model refinement. However, achieving accurate LiDAR-to-Model registration at the individual building level remains challenging, particularly due to the generalization uncertainty in semantic 3D city models at the Level of Detail 2 (LoD2). This paper addresses this gap by proposing L2M-Reg, a plane-based fine registration method that explicitly accounts for model uncertainty. L2M-Reg consists of three key steps: establishing reliable plane correspondence, building a pseudo-plane-constrained Gauss-Helmert model, and adaptively estimating vertical translation. Overall, extensive experiments on five real-world datasets demonstrate that L2M-Reg is both more accurate and computationally efficient than current leading ICP-based and plane-based methods. Therefore, L2M-Reg provides a novel building-level solution regarding LiDAR-to-Model registration when model uncertainty is present. The datasets and code for L2M-Reg can be found: https://github.com/Ziyang-Geodesy/L2M-Reg.

Figures (15)

• Figure 1: Sources of uncertainty in LoD2 building models. Model uncertainty primarily arises from the model generation process. In reality, building footprint points correspond to the wall plinth rather than the upper facade, resulting in a horizontal offset between the two (as indicated by the red dashed area). Since LoD2 models are generated directly from these footprint points, the modeled wall surfaces align geometrically with the plinth, but not with the actual facade above.
• Figure 2: Flowchart of the proposed L2M-Reg. Each plane of the input semantic LoD2 model is colored for better visualization, and the point cloud is colored by intensity.
• Figure 3: Data association illustration for LiDAR point clouds and wall surfaces. To establish correspondence, only points located within each wall’s buffer zone are retained. The resulting associations are colorized to facilitate clearer interpretation.
• Figure 4: Automated representative region localization. $N_i$ represents the input neighboring point cloud (colored by intensity), and $S_i$ represents the output representative subspace (in purple) corresponding to the building's plinth.
• Figure 5: Desired LiDAR plane segment extraction based on geometric consistency (GC) from local facade patch. Blue points indicate a rough facade patch at a local scale. Distinct planes extracted and merged using RANSAC are shown in different colors, while black highlights the plane segments extended after the GC check, representing more complete and representative structures.