A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation

Foundation models in computer vision, such as the Segment Anything Model (SAM), have demonstrated remarkable zero-shot performance in image segmentation. Leveraging these models for automated building segmentation can contribute to the efficiency of Scan-to-BIM workflows. Automatic 3D modelling has...

Full description

Saved in:
Bibliographic Details
Main Authors: M. S. A. Albadri, P. González-Cabaleiro, R. M. Túñez-Alcalde, A. Fernández, L. Díaz-Vilariño
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-2025/131/2025/isprs-archives-XLVIII-G-2025-131-2025.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849421299989872640
author M. S. A. Albadri
P. González-Cabaleiro
R. M. Túñez-Alcalde
A. Fernández
L. Díaz-Vilariño
author_facet M. S. A. Albadri
P. González-Cabaleiro
R. M. Túñez-Alcalde
A. Fernández
L. Díaz-Vilariño
author_sort M. S. A. Albadri
collection DOAJ
description Foundation models in computer vision, such as the Segment Anything Model (SAM), have demonstrated remarkable zero-shot performance in image segmentation. Leveraging these models for automated building segmentation can contribute to the efficiency of Scan-to-BIM workflows. Automatic 3D modelling has become widely relied on point cloud data; however, the nature of this data hinders the direct application of the foundation models. This study explores the potential use of SAM for automatic point cloud segmentation, proposing a SAM-based approach for segmenting building components, such as rooms, doors, and windows. The proposed method employs SAM to generate masks for an image that represents projected point clouds. Point clouds are then retrieved for each mask, which are further classified to identify building components. Room segmentation starts with the extraction of a section that defines the room boundary, followed by horizontal projection of the section. In contrast, door and window segmentation starts by projecting planes containing wall points onto their normal vectors. The experiments have been performed using three real case studies. The findings demonstrate the method's effectiveness without requiring any pretraining process, highlighting that the application of the foundation models in point cloud segmentation is a promising direction.
format Article
id doaj-art-3e2fbd19563f45c8a5ca2e0ac06ca568
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-3e2fbd19563f45c8a5ca2e0ac06ca5682025-08-20T03:31:30ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342025-07-01XLVIII-G-202513113810.5194/isprs-archives-XLVIII-G-2025-131-2025A SAM-Based Approach for Automatic Indoor Point Cloud SegmentationM. S. A. Albadri0P. González-Cabaleiro1R. M. Túñez-Alcalde2A. Fernández3L. Díaz-Vilariño4CINTECX, Universidade de Vigo, GeoTECH group, 36310 Vigo, SpainCINTECX, Universidade de Vigo, GeoTECH group, 36310 Vigo, SpainCINTECX, Universidade de Vigo, GeoTECH group, 36310 Vigo, SpainCINTECX, Universidade de Vigo, GeoTECH group, 36310 Vigo, SpainCINTECX, Universidade de Vigo, GeoTECH group, 36310 Vigo, SpainFoundation models in computer vision, such as the Segment Anything Model (SAM), have demonstrated remarkable zero-shot performance in image segmentation. Leveraging these models for automated building segmentation can contribute to the efficiency of Scan-to-BIM workflows. Automatic 3D modelling has become widely relied on point cloud data; however, the nature of this data hinders the direct application of the foundation models. This study explores the potential use of SAM for automatic point cloud segmentation, proposing a SAM-based approach for segmenting building components, such as rooms, doors, and windows. The proposed method employs SAM to generate masks for an image that represents projected point clouds. Point clouds are then retrieved for each mask, which are further classified to identify building components. Room segmentation starts with the extraction of a section that defines the room boundary, followed by horizontal projection of the section. In contrast, door and window segmentation starts by projecting planes containing wall points onto their normal vectors. The experiments have been performed using three real case studies. The findings demonstrate the method's effectiveness without requiring any pretraining process, highlighting that the application of the foundation models in point cloud segmentation is a promising direction.https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/131/2025/isprs-archives-XLVIII-G-2025-131-2025.pdf
spellingShingle M. S. A. Albadri
P. González-Cabaleiro
R. M. Túñez-Alcalde
A. Fernández
L. Díaz-Vilariño
A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
title A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
title_full A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
title_fullStr A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
title_full_unstemmed A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
title_short A SAM-Based Approach for Automatic Indoor Point Cloud Segmentation
title_sort sam based approach for automatic indoor point cloud segmentation
url https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/131/2025/isprs-archives-XLVIII-G-2025-131-2025.pdf
work_keys_str_mv AT msaalbadri asambasedapproachforautomaticindoorpointcloudsegmentation
AT pgonzalezcabaleiro asambasedapproachforautomaticindoorpointcloudsegmentation
AT rmtunezalcalde asambasedapproachforautomaticindoorpointcloudsegmentation
AT afernandez asambasedapproachforautomaticindoorpointcloudsegmentation
AT ldiazvilarino asambasedapproachforautomaticindoorpointcloudsegmentation
AT msaalbadri sambasedapproachforautomaticindoorpointcloudsegmentation
AT pgonzalezcabaleiro sambasedapproachforautomaticindoorpointcloudsegmentation
AT rmtunezalcalde sambasedapproachforautomaticindoorpointcloudsegmentation
AT afernandez sambasedapproachforautomaticindoorpointcloudsegmentation
AT ldiazvilarino sambasedapproachforautomaticindoorpointcloudsegmentation