Roof plane segmentation by combining multiple images and point clouds

authored by: Franz Rottensteiner
Abstract: A new method for roof plane detection using multiple aerial images and a point cloud is presented. It takes advantage of the fact that segmentation results for different views look different even if the same parameters are used for the original segmentation algorithm. The point cloud can be generated by image matching or by airborne laserscanning. Plane detection starts by a segmentation that is applied to each of the images. The point cloud is used to determine which image segments correspond to planes. The best plane according to a criterion is selected and matched with segments in the other images. Matching of segments requires a DSM generated from the point cloud, and it takes into account the occlusions in each image. This procedure is repeated until no more planes can be found. After that, planar segments are extracted based on region growing in the point cloud in areas of severe under-segmentation, and the multiple-image segmentation procedure is repeated. Finally, neighbouring regions found to be co-planar are merged. First results are presented for test site with up to nine-fold overlap. Our tests show that the method can deliver a good separation of roof planes under difficult circumstances, though the level of detail that can be achieved is limited by the resolution of the point cloud.
Organisation(s): Institute of Photogrammetry and GeoInformation (IPI)
Type: Conference article
Journal: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
Volume: 38
Pages: 245-250
No. of pages: 6
ISSN: 1682-1750
Publication date: 2010
Publication status: Published
Peer reviewed: Yes
ASJC Scopus subject areas: Information Systems, Geography, Planning and Development
Electronic version(s): https://doi.org/10.15488/1140 (Access: Unknown)

BibTeX

@article{367955a550834ed2a6c1007337f807b4,
title = "Roof plane segmentation by combining multiple images and point clouds",
abstract = "A new method for roof plane detection using multiple aerial images and a point cloud is presented. It takes advantage of the fact that segmentation results for different views look different even if the same parameters are used for the original segmentation algorithm. The point cloud can be generated by image matching or by airborne laserscanning. Plane detection starts by a segmentation that is applied to each of the images. The point cloud is used to determine which image segments correspond to planes. The best plane according to a criterion is selected and matched with segments in the other images. Matching of segments requires a DSM generated from the point cloud, and it takes into account the occlusions in each image. This procedure is repeated until no more planes can be found. After that, planar segments are extracted based on region growing in the point cloud in areas of severe under-segmentation, and the multiple-image segmentation procedure is repeated. Finally, neighbouring regions found to be co-planar are merged. First results are presented for test site with up to nine-fold overlap. Our tests show that the method can deliver a good separation of roof planes under difficult circumstances, though the level of detail that can be achieved is limited by the resolution of the point cloud.",
keywords = "Buildings, City Models, Data Fusion, Modelling, Segmentation",
author = "Franz Rottensteiner",
year = "2010",
doi = "10.15488/1140",
language = "English",
volume = "38",
pages = "245--250",
journal = "International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives",
issn = "1682-1750",
publisher = "International Society for Photogrammetry and Remote Sensing",
note = "ISPRS Technical Commission III Symposium on Photogrammetric Computer Vision and Image Analysis, PCV 2010 ; Conference date: 01-09-2010 Through 03-09-2010",
}