MLS2LoD3: Refining low LoDs building models with MLS point clouds to reconstruct semantic LoD3 building models

TL;DR

The paper tackles the challenge of scalable semantic LoD3 building reconstruction by leveraging ubiquitous LoD1/LoD2 geometry as priors and high accuracy MLS point clouds. It proposes a conflict driven refinement workflow that identifies missing facade elements through visibility analysis and resolves them via Bayesian fusion with a predefined library, followed by embedding the results into CityGML 2.0. Key contributions include the refinement from LoD1/2 to LoD3, practical guidelines for reconstructing three level facade elements, and CityGML embedding procedures demonstrated on real datasets. The work enables at scale LoD3 reconstruction with preserved cadastre identifiers, unlocking broad applications in urban planning, navigation, energy, safety, and beyond, while providing actionable guidelines to researchers and practitioners.

Abstract

Although highly-detailed LoD3 building models reveal great potential in various applications, they have yet to be available. The primary challenges in creating such models concern not only automatic detection and reconstruction but also standard-consistent modeling. In this paper, we introduce a novel refinement strategy enabling LoD3 reconstruction by leveraging the ubiquity of lower LoD building models and the accuracy of MLS point clouds. Such a strategy promises at-scale LoD3 reconstruction and unlocks LoD3 applications, which we also describe and illustrate in this paper. Additionally, we present guidelines for reconstructing LoD3 facade elements and their embedding into the CityGML standard model, disseminating gained knowledge to academics and professionals. We believe that our method can foster development of LoD3 reconstruction algorithms and subsequently enable their wider adoption.

Figures (10)

• Figure 1: Applications of semantic 3D building models at 3
• Figure 2: Refinement strategy workflow
• Figure 3: Concept of the refinement: Wall Surface serves as a projection plane for facade elements and as a link to building entity and subsequently city model.
• Figure 4: Conflicts understanding performed using combination of visibility analysis, semantically segmented point clouds and images. Adapted from wysocki2023scan2lod3.
• Figure 5: Visibility analysis employed on a voxel grid to identify 3 objects, absent in 1 and 2 building models (conflicts). a) Ray casting from sensor origin provides voxel state empty if the observation ray traverses it; occupied when it contains hit point; unknown if unmeasured; b) Joint analysis of rays and vector models provides another set of states: confirmed when occupied voxel intersects with vector plane; and conflicted when the plane intersects with an empty voxel. Adapted from wysocki2023scan2lod3.