International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
    
3D BUILDING CHANGE DETECTION USING HIGH RESOLUTION STEREO IMAGES 
AND A GIS DATABASE 
G. R. Dini!, K. Jacobsen!, F. Rottensteiner!, M. Al Rajhi!”, C Heipke' 
1 Institute of Photogrammetry and GeoInformation, Leibniz University Hannover, 
dini@ipi.uni-hannover.de 
2 Ministry of Municipal and Rural Affairs (MoMRA), Riyadh, KSA, mnalrajhi@yahoo.com 
Commission VII, WG VII/S 
KEY WORDS: Three-dimensional, Building, Change Detection, HR Satellite Images, Multitemporal, Urban Region 
ABSTRACT 
In this paper, a workflow is proposed to detect 3D building changes in urban and sub-urban areas using high-resolution stereoscopic 
satellite images of different epochs and a GIS database. Semi-global matching (SGM) is used to derive Digital Surface Models 
(DSM) and subsequently normalised digital surface models (nDSM, the difference of a DSM and a digital elevation model (DEM)), 
from the stereo pairs at each epoch. Large differences between the two DSMs are assumed to represent height changes. In order to 
reduce the effect of matching errors, heights in the nDSM of at least one epoch must also lie above a certain threshold in order to be 
considered as candidates for building change. A GIS database is used to check the existence of buildings at epoch 1. As a result of 
geometric discrepancies during data acquisition caused by different view directions and illumination conditions, the outlines of 
existing buildings do not necessarily match even in non-changed areas. Consequently, in the change map, there are streaking-shaped 
structures along the building outlines which do not correspond to actual changes. To eliminate these effects morphologic filtering is 
applied. The mask we use operates as a threshold on the shape and size of detected new blobs and effectively removes small objects 
such as cars, small trees and salt and pepper noise. The results of the proposed algorithm using IKONOS and GeoEye images 
demonstrate its performance for detecting 3D building changes and to extract building boundaries. 
1. INTRODUCTION 
1.4 Introduction 
Monitoring and managing urban sprawl is of crucial importance 
in regional planning and is an important pre-requisite for 
sustainable development of urban environments. Buildings are 
the most important features in urban and suburban areas, thus 
automatic monitoring of building change, especially in damage 
assessment and disaster situations, has recently received 
attention in photogrammetry and remote sensing. Optical 
images provide a wide range of information for detection of 
building changes either directly (spatial, spectral, radiometric 
and temporal information) or indirectly through height 
information generated from stereo images using image 
matching. High resolution stereo images from space provide 
appropriate tools to detect changes in residential areas (Im and 
Jensen, 2005),(Zhang and Gruen, 2006). 
A wide range of image processing and computer vision 
techniques including spectral indices (e.g., NDVI), geometric 
(e.g., shape and size) and height information (e.g. DSM, 
differential DSMs) have been developed for change detection 
based on remotely sensed images and geospatial databases (Im 
and Jensen, 2005), (Chaabouni-Chouayakh and Reinartz, 2011), 
(Bouziani et al., 2010), Champion (2007). There are two main 
strategies which are considered in change detection algorithms 
using remote sensing data: change enhancement emphasises 
image differences without any information about the type of 
    
change (i.e, changed and un-changed pixels). The second 
approach is known as ‘from-to’ strategy that monitors the 
land use changes during a period of time by determining the 
change from land use A to land use B (Im and Jensen, 2005). 
In this paper, the first approach along with additional 
refinement is applied for stereo images of different epochs in 
order to detect building changes based on height changes. A 
GIS database containing building footprints can optionally be 
used to enhance the results. We do not aim at accurately 
delineating buildings outlines, instead the proposed approach is 
meant to act as a reliable alarm system for detecting changes, 
while delineation is assumed to be performed manually at a 
later stage. In the next section a short review of the literature is 
given. Section 2 describes the type of data used in our study. 
Section 3 contains a description of our approach. Experimental 
results are reported in section 4, section 5 concludes the paper. 
This paper continues and extends the work of Alobeid et al. 
(2011). 
1.2 State-of-the-art 
Change detection using high resolution stereoscopic images is 
an important task in updating urban areas. Champion (2007) 
matches high resolution aerial images for generating a DSM in 
order to update the building layer of a 2D cadastral database. 
He subdivided the detection of changes into two main steps, 
automatic verification of the database and detection of new 
buildings. In the first step, all buildings of the database are