Confidence-Driven Facade Refinement of 3D Building Models Using MLS Point Clouds

Abstract

Digital twins require continuous maintenance to meet the increasing demand for high-precision geospatial data. However, traditional coarse CityGML building models, typically derived from Airborne Laser Scanning (ALS), often exhibit significant geometric deficiencies, particularly regarding facade accuracy due to the nadir perspective of airborne sensors. Integrating these coarse models with high-precision Mobile Laser Scanning (MLS) data is essential to recover detailed facade geometry. Unlike reconstruction-from-scratch approaches that discard existing semantic information and rely heavily on complete data coverage, this work presents an automated refinement framework that utilizes the coarse model as a geometric prior. This method enables targeted updates to facade geometry even in complex urban environments. It integrates surface matching to identify outdated surfaces and employs a binary integer optimization to select optimal faces from candidate data. Crucially, hard constraints are enforced within the optimization to ensure the topological validity of the refined output. Experimental results demonstrate that the proposed approach effectively corrects facade misalignments, reducing the Cloud-to-Mesh RMSE by approximately 36% and achieving centimeter-level alignment. Furthermore, the framework guarantees strictly watertight and manifold geometry, providing a robust solution for upgrading ALS-derived city models.

Figures (7)

• Figure 1: Overview of the proposed refinement workflow. The pipeline takes a coarse model and a point cloud as inputs, performing: (1) model to surface matching to extract facade info; (2) candidate face generation; (3) coverage confidence computation; and (4) an optimization-based face selection that guarantees a watertight and manifold refined model.
• Figure 2: Overview of the coverage measuring metrics. The figure illustrates the four key metrics used to evaluate the geometric alignment between a candidate face $s_c$ and a reference plane $s_r$. (a) Angle filtering, which compares the normal vectors $N_c$, $N_r$, (b) Distance filtering, based on the distance $d_i$ of point $p_i$ to the plane, (c) Bounding box overlap, measuring the volume intersection of bounding boxes $B_1$, $B_2$, (d) Projection area overlap, which measures the 2D area of overlap.
• Figure 3: Overview of the study area in Hildesheim, Germany. (a) The spatial extent of the CityGML building models covering the city scale. (b) The geographic context within the city, with the red box indicating the location of the selected test site. (c) Detailed view of the 300 $\times$ 300 m test site used for experiments. (Basemap: OpenStreetMap)
• Figure 4: Large-scale refinement results on the Hildesheim dataset (approx. 300 m × 300 m). (a) displays the initial coarse models, while (b) shows the refined building models. (c) (d) The original and refined histogram of C2M distances. (e) The boxplot of the distances. The red line represents the median value. The plot compares the error distribution spread of the initial input coarse meshes (blue) versus the refined models (orange).
• Figure 5: Parameter sensitivity analysis on the coverage threshold $\tau_{cov}$. The plot evaluates the average C2M errors (left axis) and the topological validity rate (right axis) across varying thresholds.