Please use this identifier to cite or link to this item: http://theses.ncl.ac.uk/jspui/handle/10443/3829
Title: Automatic reconstruction of three-dimensional building models from dense image matching datasets
Authors: McClune, Andrew Philip
Issue Date: 2017
Publisher: Newcastle University
Abstract: The generation of three-dimensional (3D) building models without roof geometry is currently easily automated using a building footprint and single height value. The automatic reconstruction of roof structures, however, remains challenging because of the complexity and variability in building geometry. Attempts from imagery have utilised high spatial resolution but have only reconstructed simple geometry. This research addresses the complexity of roof geometry reconstruction by developing an approach, which focuses on the extraction of corners to reconstruct 3D buildings as boundary representation models, to try overcome the limitations of planar fitting procedures, which are currently favoured. Roof geometry information was extracted from surface models, true orthophotos and photogrammetric point clouds; reconstructed at the same spatial resolution of the captured aerial imagery, with developments in pixel-to-pixel matching. Edges of roof planes were extracted by the Canny edge detector, and then refined with a workflow based on the principles of scan-line segmentation to remove false positive detection. Line tracing procedures defined the corner positions of the extracted edges. A connectivity ruleset was developed, which searches around the endpoints of unconnected lines, testing for potential connecting corners. All unconnected lines were then removed reconstruct 3D models as a closed network of connecting roof corners. Building models have been reconstructed both as block models and also with roof structures. The methodology was tested on data of Newcastle upon Tyne, United Kingdom, with results showing corner extraction success at 75% and to within a planimetric accuracy of ±0.5 m. The methodology was then tested on data of Vaihingen, Germany, which forms part of the ISPRS 3D reconstruction benchmark. This allowed direct comparisons to be made with other methods. The results from both study areas showed similar planimetric accuracy of extracted corners. However, both sites were not as successful in the reconstruction of roof planes.
Description: PhD Thesis
URI: http://hdl.handle.net/10443/3829
Appears in Collections:School of Civil Engineering and Geosciences

Files in This Item:
File Description SizeFormat 
McClune, A.M 2017.pdfThesis8.19 MBAdobe PDFView/Open
dspacelicence.pdfLicence43.82 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.