2008
983
DESIGN AND DEVELOPMENT OF A NEW MODEL FOR AUTOMATIC CHANGE
DETECTION OF BUILDINGS FROM AERIAL IMAGES
Simintaj ziraksaz ,Hamid Ebadi 2 ,Farshid Famood Ahmadi 3 , saeed sadeghian 4
simin_ziraksaz@yahoo.comM.Sc. student, E-mail: 1.
ebadi@kntu.ac.ir Assistant Professor, E-mail:2.
farshid_famood@yahoo.com 3.Ph.D student, E-mail:
Department of Photogrammetry and Remote Sensing
K.N. Toosi university,Faculty of Geodesy and Geomatics Engineering
4.Ph.D, E-mail: sedeghian@ncc.neda.net.ir
National Cartographic Center of Iran (NCC)
Commission VI, WG VI/4
KEY WORDS: Aerial photos, Building, Change detection, Gis, Matching, Spatial database
ABSTRACT:
One of the most important tools in order to model real world phenomena is Geographic Information System. Real world phenomena
are changing continuously, so it is necessary to detect and introduce changes to system in order to update GIS databases and model
real world correctly. Geometric stability and high spatial resolution of aerial images leads to detect changes precisely. It is necessary
to consider that feature recognition and extraction processes are two important stages in all of the change detection methods in which
accurate results in detecting geometrical changes depends on performance of theses two stages. Each of these stages needs to use
image processing algorithms. On the other hand, these algorithms are complicated, time consuming and the performance of them
depends on specific conditions such as image acquisition conditions and so on. Therefore using the methods with low dependency on
image processing can reduce these problems and increase the accuracy and reliability of the results. To achieve this aim, the
integration of aerial photos and GIS spatial databases is suggested. In this research, a new algorithm was designed and implemented
for automatic change detection of buildings based on the development of least squared matching technique. The accuracy assessment
showed that change percentage of the regions that the algorithm can detect their changes correctly is 70% and the precision of this
novel approach to approximate extracted edges with the real ones is 0.45 pixels.
1. INTRODUCTION
In photogrammetry and remote sensing, matching can be
defined as the establishment of the correspondence between
various data sets (Geodaetischas seminar ss/2000). The
matching problem is also referred to as the correspondence
problem. The data sets can represent images, but also maps, or
object models and GIS data. The first step of matching is
primitive extraction. The distinction between different matching
primitives is probably the most prominent difference between
the various matching algorithms. The primitives fall into two
broad categories: either windows composed of grey values or
features extracted in each image a priori are used in the actual
matching step. The resulting algorithms are usually called: Area
based matching (ABM), and Feature based matching (FBM),
respectively (Bohuslav, 2004). Besides of these two algorithms,
the combination algorithm can be also considered. This
algorithm combines ABM and FBM in order to take the
advantages of these two algorithms (Gruen, 1985).
Correlation coefficient and least squares matching (LSM) are
two approaches used in ABM. Cross-correlation works fast and
well, if patches to be matched contain enough signal without too
much high frequency content (noise) and if geometrical and
radiometric distortions are kept at minimum. Both conditions
are often not encountered in or with aerial images, while LSM
minimizes differences in grey values between the template and
search image patches in an adjustment process where geometric
and radiometric corrections of one of matching windows are
determined (Bohuslav, 2004). On the other hand FBM is
implemented in three ways point-based matching edge-based
matching and region-based matching based on the type of
features extracted from images (Geodaetischas seminar
ss/2000). Among of these approaches edge-based matching can
presented accurate and reliable results because of its specific
radiometric characteristics (Agouris, 1992), (Baltsavias, 1991).
So the combination of LSM and edge-based matching is used in
this article as an optimal method to obtain desired results in
building change detection process.
The goal of our study is to develop LSM for the identification
of changes in buildings outlines. In this paper change is
extracted thorough comparisons of observations. The
differences in general exposure conditions among two different
images in two distinct time instances may affect substantially
the performance of the above described matching method. In
order to minimize the effects of variations on our solution, we
have to allow edge pixels to influence the solution more than
the rest of the template. This can be performed by manipulating
the corresponding weight matrix P.
Therefore our approach is based on the use of least squares
template matching, where prior data are analyzed to provide
template information. The product of such a process is the
identification of changes in object outlines. Our work is
innovative in its use of prior information to provide templates
for matching, and in its analysis of template information to
assign proper weights in the least squares solution. In this paper
we present theoretical models and implementation
considerations behind our approach for change detection. The
steps of implementation of this proposed algorithm is described
as below.