Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SimpliCity: Reconstructing Buildings with Simple Regularized 3D Models

Jean-Philippe Bauchet, Raphael Sulzer, Florent Lafarge, Yuliya Tarabalka

TL;DR

SimpliCity addresses the need for simple, regularized 3D building meshes reconstructed from airborne LiDAR. It introduces a two-stage planimetric framework: (i) regularizing a 2D polygonal partition derived from detected planes, and (ii) extruding this partition to 3D via optimization that enforces roof planarity and preserves vertical discontinuities and horizontal rooftop edges. The method yields watertight, 2-manifold, low-complexity meshes with fidelity comparable to state-of-the-art and substantially reduced vertex/facets counts, while maintaining robust geometric guarantees. This approach enables scalable city-scale reconstruction and practical applications such as texture mapping, with potential extensions to higher levels of detail and web-based deployment.

Abstract

Automatic methods for reconstructing buildings from airborne LiDAR point clouds focus on producing accurate 3D models in a fast and scalable manner, but they overlook the problem of delivering simple and regularized models to practitioners. As a result, output meshes often suffer from connectivity approximations around corners with either the presence of multiple vertices and tiny facets, or the necessity to break the planarity constraint on roof sections and facade components. We propose a 2D planimetric arrangement-based framework to address this problem. We first regularize, not the 3D planes as commonly done in the literature, but a 2D polyhedral partition constructed from the planes. Second, we extrude this partition to 3D by an optimization process that guarantees the planarity of the roof sections as well as the preservation of the vertical discontinuities and horizontal rooftop edges. We show the benefits of our approach against existing methods by producing simpler 3D models while offering a similar fidelity and efficiency.

SimpliCity: Reconstructing Buildings with Simple Regularized 3D Models

TL;DR

SimpliCity addresses the need for simple, regularized 3D building meshes reconstructed from airborne LiDAR. It introduces a two-stage planimetric framework: (i) regularizing a 2D polygonal partition derived from detected planes, and (ii) extruding this partition to 3D via optimization that enforces roof planarity and preserves vertical discontinuities and horizontal rooftop edges. The method yields watertight, 2-manifold, low-complexity meshes with fidelity comparable to state-of-the-art and substantially reduced vertex/facets counts, while maintaining robust geometric guarantees. This approach enables scalable city-scale reconstruction and practical applications such as texture mapping, with potential extensions to higher levels of detail and web-based deployment.

Abstract

Automatic methods for reconstructing buildings from airborne LiDAR point clouds focus on producing accurate 3D models in a fast and scalable manner, but they overlook the problem of delivering simple and regularized models to practitioners. As a result, output meshes often suffer from connectivity approximations around corners with either the presence of multiple vertices and tiny facets, or the necessity to break the planarity constraint on roof sections and facade components. We propose a 2D planimetric arrangement-based framework to address this problem. We first regularize, not the 3D planes as commonly done in the literature, but a 2D polyhedral partition constructed from the planes. Second, we extrude this partition to 3D by an optimization process that guarantees the planarity of the roof sections as well as the preservation of the vertical discontinuities and horizontal rooftop edges. We show the benefits of our approach against existing methods by producing simpler 3D models while offering a similar fidelity and efficiency.
Paper Structure (18 sections, 1 equation, 7 figures, 2 tables)

This paper contains 18 sections, 1 equation, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Model-driven.
  4. Mesh simplification.
  5. Plane assembly.
  6. Neural models.
  7. Proposed method
  8. Overview
  9. Construction of 2D polygonal partitions
  10. Regularization of 2D polygonal partitions
  11. Extrusion
  12. Experiments
  13. Experimental setup
  14. Results
  15. Ablation
  16. ...and 3 more sections

Figures (7)

  • Figure 1: SimpliCity. Our building reconstruction method produces a simple, regularized mesh while being a faithful approximation of the input Lidar scan. In contrast, the output meshes of Geoflow peters2022geoflow and City3D huang2022city3d contain a much higher number of facets, including tiny ones that correct connectivity approximations around corners.
  • Figure 2: Overview. Starting from a LiDAR scan and a building footprint \ref{['subfig:pipeline_a']}, we first construct a 2D polygonal partition of the roof structure (\ref{['subfig:pipeline_b']}, Section \ref{['sec-step1']}, Fig. \ref{['fig:step1']}). The partition is then regularized to both enforce orthogonality, parallelism and collinearity between edges and simplify the vertex layout (\ref{['subfig:pipeline_c']}, Section \ref{['sec-step2']}, Fig. \ref{['fig:step2']}). Finally, the partition is extruded using an optimization procedure that preserves the planarity of roof sections and the horizontality of rooftop edges (\ref{['subfig:pipeline_d']}, Section \ref{['sec-step3']}, Fig. \ref{['fig:step3']}).
  • Figure 3: Construction of 2D polygonal partition. Line-segments at the intersection of adjacent detected planes and on their boundary are projected into the horizontal plane, jointly with the input footprint line-segments \ref{['subfig:step1_a']}. An initial, dense 2D polygonal partition is built from all these line-segments by kinetic simulation \ref{['subfig:step1_b']} and enriched by plane labels (\ref{['subfig:step1_c']}, colored polygons). Polygonal cells with same label are then regrouped to form a 2D polygonal partition that describes the roof structure \ref{['subfig:step1_d']}.
  • Figure 4: Regularization of 2D polygonal partition. The 2D polygonal partition \ref{['subfig:step2_a']} is first simplified by collapsing small edges (\ref{['subfig:step2_b']}, see close-ups). The regularity graph $G$ is then built by detecting pairs of near-parallel edges (green lines) and near-orthogonal edges (purple lines) in \ref{['subfig:step2_c']}. A global optimization of the vertex coordinates constrained by $G$ is then performed to regularize the partition (\ref{['subfig:step2_d']}).
  • Figure 5: Extrusion. Each cell of the 2D regularized partition \ref{['subfig:step3_a']} is first extruded according to its associated detected plane \ref{['subfig:step3_b']}. Close 3D vertices with same x- and y-coordinates are then merged, breaking the planarity of the extruded facets \ref{['subfig:step3_c']}. The z-coordinates of the vertices are finally readjusted using a global optimization under facet planarity constraints \ref{['subfig:step3_d']}.
  • ...and 2 more figures