Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reconstructing facade details using MLS point clouds and Bag-of-Words approach

Thomas Froech, Olaf Wysocki, Ludwig Hoegner, Uwe Stilla

TL;DR

This work tackles the challenge of reconstructing detailed façade geometry from MLS point clouds without relying on rectangularity-by bounding-box assumptions. It extends the Bag-of-Words framework by linking point-cloud observations to a library of CAD façade elements and enriching descriptors with semi-global information, using ORB/HOG features and a per-model sampling strategy to build a robust codebook. The approach demonstrates improved matching accuracy over a conventional BoW setup, validated on synthetic noisy data and the TUM-Façade dataset, with Jensen–Shannon divergence often providing the best performance for real-world data. The resulting method enables more realistic façade reconstructions and has potential applications in automated driving testing and solar-potential estimation, moving beyond simple rectangular representations toward richer architectural details.

Abstract

In the reconstruction of façade elements, the identification of specific object types remains challenging and is often circumvented by rectangularity assumptions or the use of bounding boxes. We propose a new approach for the reconstruction of 3D façade details. We combine MLS point clouds and a pre-defined 3D model library using a BoW concept, which we augment by incorporating semi-global features. We conduct experiments on the models superimposed with random noise and on the TUM-FAÇADE dataset. Our method demonstrates promising results, improving the conventional BoW approach. It holds the potential to be utilized for more realistic facade reconstruction without rectangularity assumptions, which can be used in applications such as testing automated driving functions or estimating façade solar potential.

Reconstructing facade details using MLS point clouds and Bag-of-Words approach

TL;DR

This work tackles the challenge of reconstructing detailed façade geometry from MLS point clouds without relying on rectangularity-by bounding-box assumptions. It extends the Bag-of-Words framework by linking point-cloud observations to a library of CAD façade elements and enriching descriptors with semi-global information, using ORB/HOG features and a per-model sampling strategy to build a robust codebook. The approach demonstrates improved matching accuracy over a conventional BoW setup, validated on synthetic noisy data and the TUM-Façade dataset, with Jensen–Shannon divergence often providing the best performance for real-world data. The resulting method enables more realistic façade reconstructions and has potential applications in automated driving testing and solar-potential estimation, moving beyond simple rectangular representations toward richer architectural details.

Abstract

In the reconstruction of façade elements, the identification of specific object types remains challenging and is often circumvented by rectangularity assumptions or the use of bounding boxes. We propose a new approach for the reconstruction of 3D façade details. We combine MLS point clouds and a pre-defined 3D model library using a BoW concept, which we augment by incorporating semi-global features. We conduct experiments on the models superimposed with random noise and on the TUM-FAÇADE dataset. Our method demonstrates promising results, improving the conventional BoW approach. It holds the potential to be utilized for more realistic facade reconstruction without rectangularity assumptions, which can be used in applications such as testing automated driving functions or estimating façade solar potential.
Paper Structure (21 sections, 9 equations, 14 figures, 2 tables)

This paper contains 21 sections, 9 equations, 14 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. 3 Building Model Reconstruction
  4. Bag of Words Approach
  5. Methodology
  6. Overview
  7. Sampling of the CAD models
  8. Binary Image Generation and Processing
  9. Feature Extraction
  10. Incorporation of Semi-Global Features
  11. Clustering
  12. Histogram Comparison
  13. Experiments
  14. Experiments on models superimposed with random noise
  15. Experiments on the TUM-FAÇADE dataset
  16. ...and 6 more sections

Figures (14)

  • Figure 1: Overview of the approach, left side: training process, right side inference(illustration based on Memon2019)
  • Figure 2: a) Point cloud sampled from CAD b) Point cloud extracted from the TUM-Façade dataset
  • Figure 3: Exemplary image processing on a projected point cloud: a) projected image of a point cloud, b) dilated image, c) edge detection (Laplace), d) line simplification (Douglas-Peucker).
  • Figure 4: Examples for feature extraction: a) ORB-keypoints b) HOG-image
  • Figure 5: Incorporation of semi-global features
  • ...and 9 more figures