SAR SIMULATION BASED CHANGE DETECTION WITH HIGH-RESOLUTION SAR 
IMAGES IN URBAN ENVIRONMENTS 
Timo Balz 
Institute for Photogrammetry (ifp), University of Stuttgart, Germany 
Geschwister-Scholl-Strasse 24D, D-70174 Stuttgart 
timo.balz@ifp.uni-stuttgart.de 
Commission VII, WG 4 
KEY WORDS: SAR, Change Detection, City, Disaster, Radar, Rectification, Simulation 
ABSTRACT: 
Combined processing using different sensor types, i.e. for applications like change detection, requires a good geo-referencing. Fur- 
thermore the individual sensor properties have to be taken into account. SAR systems are side-looking and run-time systems. They 
suffer from occlusions and ambiguities especially in urban areas. Additionally layover and shadow effects disturb the geo-referencing 
of SAR images in urban areas, which is a prerequisite for a successful change detection. An improved geo-referencing can be 
achieved by simulating 3D-city models or street datasets using a SAR simulator and comparing the simulated image to the real 
image. Correspondences between simulated and real image can be used for geo-referencing the image according to the coordinates of 
the 3D-city model or street dataset. The geo-referenced dataset can afterwards be used for change detection analysis. SAR images 
represent a side-view of the three dimensional world. An automated change detection using SAR images should take this fact into 
consideration and therefore should use 3D-models as reference for the change-detection. These models are simulated and the 
simulated image is compared to the geo-referenced image, revealing changes between the simulated model and the real image. 
1. INTRODUCTION 
The urban environment is of the utmost importance for human 
society. In 2001, around 50% of the human population lived in 
cities and these numbers are still rising, especially in less devel- 
oped countries (UNCHS, 2001). The dense placement of build- 
ings in cities requires a good resolution of the remote sensing 
systems, to distinguish between the neighbouring buildings. 
Modern high-resolution airborne SAR systems reach very high 
resolutions up to 10cm (Ender & Brenner, 2003) and are there- 
fore useable for remote sensing applications in urban environ- 
ments. 
Using SAR has some advantages, related to the capability of 
these systems to operate at day and night and under nearly all 
weather conditions. This is most beneficial for time dependant 
applications, like disaster management. But unfortunately SAR 
has also a lot of disadvantages, especially while using it in ur- 
ban environments. These disadvantages are related to the run- 
time geometry and side-looking properties of the SAR system, 
leading to occlusions and ambiguities. In very dense urban envi- 
ronments, containing tall buildings, it is sometimes impossible 
to observe some areas at all. 
For collecting data, it may therefore, be better to use other types 
of sensors. LIDAR, for example, is more suitable for building 
reconstruction and generating city models (Haala & Brenner, 
1999), although it is possible to reconstruct buildings from 
InSAR (Sórgel, 2003). In case of an emergency, SAR is benefi- 
cial, especially for change detection applications combined with 
GIS data or remote sensing data from other sensors. 
For change detection applications using different types of sen- 
sors, the data fusion is the main problem. To solve this problem, 
the individual sensor properties of the used sensors have to be 
taken into account. In the approach described in this paper, 
available 3D-city models are simulated using a SAR simulator. 
The simulated images, afterwards, are being compared to the 
real SAR images for geo-referencing and change detection pur- 
472 
poses. The geo-referencing of the SAR data can be done using 
3D-city models or GIS street data. In this approach, changes in 
the 3D-city model are detected by comparing the SAR simu- 
lated image of the 3D-model with the real SAR image. The 
simulated image represents the expected value. Differences be- 
tween the expectations, derived from the simulation, and the 
real SAR image are supposed to be caused by changes. Al- 
though these differences may have other reasons, too. 
2. SAR PRINCIPLE 
SAR images differ from optical images not only in the wave- 
length but also in the geometrical properties. SAR images are 
run-time images and the SAR systems are side-looking systems. 
Due to this fact, it is difficult to combine SAR data and optical 
data, because of the different image geometries. Especially the 
layover and fore-shortening effects have to be taken into consid- 
eration while working on the fusion of SAR data and optical 
data. 
e, SAR 
    
BC MAD E 
far 
range 
near 
range 
Figure 1. SAR geometry in range 
  
Interi 
  
In Fi 
single 
assun 
age. I 
Care 
sition 
geom 
shadc 
shadc 
In de 
plex. 
the ir 
shad 
also « 
impo: 
build 
areas 
^N 
A SA 
image 
ments 
choos 
occlu: 
For tl 
ment. 
not b 
standi 
simul 
helpfi