elib
DLR-Header
DLR-Logo -> http://www.dlr.de
DLR Portal Home | Imprint | Privacy Policy | Contact | Deutsch
Fontsize: [-] Text [+]

Spectral Information Retrieval for Sub-Pixel Building Detection

Avbelj, Janja (2012) Spectral Information Retrieval for Sub-Pixel Building Detection. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, I-7 , pp. 61-66. Copernicus Publications. XXII ISPRS Congress 2012, 25. Aug. – 01. Sep. 2012, Melbourne, Australia. ISSN doi:10.5194/isprsannals-I-7-61-2012

Full text not available from this repository.

Official URL: http://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/I-7/61/2012/isprsannals-I-7-61-2012.html

Abstract

Building extraction from imagery has been an active research area for decades. However, the precise building detection from hyperspectral (HSI) images solely is a less often addressed research question due to the low spatial resolution of data. The building boundaries are usually represented by spectrally mixed pixels, and classical edge detector algorithms fail to detect borders with sufficient completeness. The idea of the proposed method is to use fraction of materials in mixed pixels to derive weights for adjusting building boundaries. The building regions are detected using seeded region growing and merging in a HSI image; for the initial seed point selection the digital surface model (DSM) is used. Prior to region growing, the seeds are statistically tested for outliers on the basis of their spectral characteristics. Then, the border pixels of building regions are compared in spectrum to the seed points by calculating spectral dissimilarity. From this spectral dissimilarity the weights for weighted and constrained least squares (LS) adjustment are derived. We used the Spectral Angle Mapper (SAM) for spectral similarity measure, but the proposed boundary estimation method could benefit from soft classification or spectral unmixing results. The method was tested on a HSI image with spatial resolution of 4 m, and buildings of rectangular shape. The importance of constraints to the relations between building parts, e.g. perpendicularity is shown on example with a building with inner yards. The adjusted building boundaries are compared to the laser DSM, and have a relative accuracy of boundaries 1/4 of a pixel.

Item URL in elib:https://elib.dlr.de/77567/
Document Type:Conference or Workshop Item (Speech, Paper)
Additional Information:Young Author Award
Title:Spectral Information Retrieval for Sub-Pixel Building Detection
Authors:
AuthorsInstitution or Email of AuthorsAuthors ORCID iD
Avbelj, JanjaJanja.Avbelj (at) dlr.de und TU MünchenUNSPECIFIED
Date:16 July 2012
Journal or Publication Title:ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Refereed publication:Yes
Open Access:Yes
Gold Open Access:Yes
In SCOPUS:No
In ISI Web of Science:No
Volume:I-7
Page Range:pp. 61-66
Editors:
EditorsEmail
Shortis, M.UNSPECIFIED
Wagner, W.UNSPECIFIED
Hyppä, J.UNSPECIFIED
Publisher:Copernicus Publications
Series Name:ISPRS Annals
ISSN:doi:10.5194/isprsannals-I-7-61-2012
Status:Published
Keywords:Hyper spectral, DEM/DTM, Edge, Building, Urban, Photogrammetry
Event Title:XXII ISPRS Congress 2012
Event Location:Melbourne, Australia
Event Type:international Conference
Event Dates:25. Aug. – 01. Sep. 2012
Organizer:International Society of Photogrammetry and Remote Sensing
HGF - Research field:Aeronautics, Space and Transport
HGF - Program:Space
HGF - Program Themes:Earth Observation
DLR - Research area:Raumfahrt
DLR - Program:R EO - Erdbeobachtung
DLR - Research theme (Project):R - Vorhaben hochauflösende Fernerkundungsverfahren
Location: Oberpfaffenhofen
Institutes and Institutions:Remote Sensing Technology Institute > Photogrammetry and Image Analysis
Deposited By:INVALID USER
Deposited On:15 Oct 2012 07:44
Last Modified:15 Oct 2012 07:44

Repository Staff Only: item control page

Browse
Search
Help & Contact
Information
electronic library is running on EPrints 3.3.12
Copyright © 2008-2017 German Aerospace Center (DLR). All rights reserved.