3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points

Real-world 3D reconstructions of building façades in LoD3 and beyond are not yet widely available on the mass market due to financial and technological barriers, as well as the challenges of automated modeling. We propose a novel method for extracting 3D façade elements using i...

Full description

Saved in:
Bibliographic Details
Main Authors: F. Frank, V. Shah, S. Worbis, L. Hoegner
Format: Article
Language:English
Published: Copernicus Publications 2025-07-01
Series:The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Online Access:https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/473/2025/isprs-archives-XLVIII-G-2025-473-2025.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849245789751083008
author F. Frank
V. Shah
S. Worbis
L. Hoegner
author_facet F. Frank
V. Shah
S. Worbis
L. Hoegner
author_sort F. Frank
collection DOAJ
description Real-world 3D reconstructions of building façades in LoD3 and beyond are not yet widely available on the mass market due to financial and technological barriers, as well as the challenges of automated modeling. We propose a novel method for extracting 3D façade elements using image-based instance segmentation and scale-invariant object contour points (SIOCP). Our methodology uses RGB images, camera parameters, absolute 6DoF pose and position, as well as LoD2 building information. The images are processed using instance segmentation with YOLOv8 and SAM, complemented by classical and enhanced algorithms for line and edge detection. The SIOCP method refines object contour lines from instance segmentation by incorporating LoD2 building data and 6DoF information. Subsequently, the keypoints are estimated and the single-camera image 6DoF pose is reconstructed using a PnP solver. From these 6DoF poses a photogrammetrically point cloud is generated, and semantically- and instance-segmented with SuperCluster. The segmentation results are intended for future comparisons with other point clouds and LoD3 reconstructions. The presented approach is still under development, so the current results are limited. In summary, this paper introduces a key component of our vision for LoD3 reconstruction by using handheld devices.
format Article
id doaj-art-4a3e1abb320340ccafc75916acfed07a
institution Kabale University
issn 1682-1750
2194-9034
language English
publishDate 2025-07-01
publisher Copernicus Publications
record_format Article
series The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
spelling doaj-art-4a3e1abb320340ccafc75916acfed07a2025-08-20T03:58:41ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342025-07-01XLVIII-G-202547348010.5194/isprs-archives-XLVIII-G-2025-473-20253D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour PointsF. Frank0V. Shah1S. Worbis2L. Hoegner3IWT – Institute of Economic and Technical Sciences, Friedrichshafen, GermanyIWT – Institute of Economic and Technical Sciences, Friedrichshafen, GermanyDepartment of Geoinformatics, Hochschule München University of Applied Sciences, Munich, GermanyDepartment of Geoinformatics, Hochschule München University of Applied Sciences, Munich, GermanyReal-world 3D reconstructions of building façades in LoD3 and beyond are not yet widely available on the mass market due to financial and technological barriers, as well as the challenges of automated modeling. We propose a novel method for extracting 3D façade elements using image-based instance segmentation and scale-invariant object contour points (SIOCP). Our methodology uses RGB images, camera parameters, absolute 6DoF pose and position, as well as LoD2 building information. The images are processed using instance segmentation with YOLOv8 and SAM, complemented by classical and enhanced algorithms for line and edge detection. The SIOCP method refines object contour lines from instance segmentation by incorporating LoD2 building data and 6DoF information. Subsequently, the keypoints are estimated and the single-camera image 6DoF pose is reconstructed using a PnP solver. From these 6DoF poses a photogrammetrically point cloud is generated, and semantically- and instance-segmented with SuperCluster. The segmentation results are intended for future comparisons with other point clouds and LoD3 reconstructions. The presented approach is still under development, so the current results are limited. In summary, this paper introduces a key component of our vision for LoD3 reconstruction by using handheld devices.https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/473/2025/isprs-archives-XLVIII-G-2025-473-2025.pdf
spellingShingle F. Frank
V. Shah
S. Worbis
L. Hoegner
3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
title 3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
title_full 3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
title_fullStr 3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
title_full_unstemmed 3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
title_short 3D Façade Element Extraction from Image-based Instance Segmentation and Scale-Invariant Object Contour Points
title_sort 3d facade element extraction from image based instance segmentation and scale invariant object contour points
url https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/473/2025/isprs-archives-XLVIII-G-2025-473-2025.pdf
work_keys_str_mv AT ffrank 3dfacadeelementextractionfromimagebasedinstancesegmentationandscaleinvariantobjectcontourpoints
AT vshah 3dfacadeelementextractionfromimagebasedinstancesegmentationandscaleinvariantobjectcontourpoints
AT sworbis 3dfacadeelementextractionfromimagebasedinstancesegmentationandscaleinvariantobjectcontourpoints
AT lhoegner 3dfacadeelementextractionfromimagebasedinstancesegmentationandscaleinvariantobjectcontourpoints