\N AREAS
va 215-0004,
lagatsuta-cho,
ly observation of
is is promising.
automatic image
e of no available
fore and after the
only 2D changes’
disaster. We call
: between images
ch candidates are
constrained by a
lese processes y-
ure, formation of
rea, the adaptive
the image pair is
ping process.
additional terrain
ion are available.
d 3D matching is
veral years later
erpretation. For
ility of detection
re about 30 % in
s approach as 2D
proach aims at
tribution but also
g digital terrain
ore and after the
ethod.
stereo model is
nage registration
etecting changed
t is based on the
: (Kosugi et al.
The one of the
ocal deformation.
nt matching in
tching method is
o the additional
created before
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
disaster in local government. In this case, it is assumed that
stereoscopic aerial photos and exterior orientation are available
by GPS/IMU. Former digital terrain data such as DSM or
digital maps is transformed based on exterior orientation
parameters and 3D matching. Changed areas are detected as
inconsistent texture or height anomaly.
Evaluation tests were performed with actual aerial photos that
were taken right after of the earthquake and several years later.
2. CONCEPT AND STATISTICS
This study is objected for timely observation of area damaged
by earthquake, especially for buildings, by establishing
automatic change detection system. Automatic change
detection system is needed for a disaster area in its carly stage
because such information is demanded for planning rescue
mission or limiting the expansion of the damage. Fig.l shows
the flow of urgent correspondence system for a huge earthquake
occurrence, which is the target of this study. In this system,
there are two typical cases of change detection approaches. The
first one is a direct comparison method and the other one is an
indirect comparison method respectively. :
The assumptions for the two approaches are as follows.
(1) For direct comparison method
1) No orientation parameters for aerial photos before and after
the earthquake are available except approximate geographic
coordinate.
(2) For indirect comparison method
1) Stereo pairs of aerial imageries before and after the
earthquake are available.
2) Internal and external orientation parameters are known.
3) 2D or 2.5D digital map data for buildings of target area are
prepared in local government's spatiotemporal GIS
(Geographic Information System) data server and are
available for the change detection process.
: sitioning a E : :
: Feste vim and B> Change detection B» Mapping
: sn Digital image sue” *
: : = Tr
: \ EE ee À
i — — i is nat Poin ID image matchisg Texture analysis
: ith orientali ;
5 Digital stereo aerial photo
(with GPSAMU data)
% ere
x Image 4
- transmission 2
Imagery of
Detected
changes
Di Government, |
Bnsier Emergency
map administration
Mapping by using
spatio-temporal GIS
Digital image by
: a helicopter
+ (with GPS/IMU data)
im 3
“satellite i 1
WERT >
Mu =
“BQ I phi i Spatio-temporal GIS
GPSEmth ¥.. 4: in local gove rnment
Observation Network
Figure 1. Urgent correspondence system for a huge
earthquake occurrence
External orientation parameters are obtained by on flight
analysis of GPS/IMU.
In the first approach, we can use only single aerial imagery
before and after the earthquake for the target area, which also
means that no stereo pairs are available and 3D information is
not acquired. In this case, change detection is executed by 2D
comparison of images. We call this approach 2D image
matching method and this process is realized with a unique
registration method called adaptive nonlinear mapping in this
study.
On the other hand, indirect comparison method utilizes stereo
pair imageries and orientation parameters in addition to digital
map data. The changes of disaster area are detected as 3D shift.
3. CHANGE DETECTION BY 2D IMAGE MATCHING
METHOD
3.1 Process Flow
Fig.2 shows the process flow of 2D image matching method. In
the first step, a pair of aerial imageries before and after the
change is used to form an imaginary stereo model, which is
objected for facilitating and stabilizing image matching process.
In the next step, image matching is carried out. Geometrical
registration processes such as affine transformation or Helmert's
transformation are not applicable or sufficient for this process
due to the influence of parallax. To solve this problem,
adaptive nonlinear mapping method is applied in this study. In
the last step, change area is detected as the inconsistent area of
adaptive nonlinear mapping. The details of the processes are
described in the following sections.
Formation of imaginary
stereo mode
y
Image matching by
adaptive nonlinear mapping
Detection of change area
Figure 2. Process flow of 2D image matching method
3.2 Formation of Imaginary Stereo Model
Formation of imaginary stereo model is for reducing calculation
time and increasing the stability of adaptive nonlinear mapping
process. Fig.3 shows the processing flow for formation of
imaginary stereo model. In the first step, pre-processing is
conducted such as contrast adjustment by automatic process or
adjustment of image resolution by manual process and so on.
The next step detects matching points in a pair of images taken
at different time. Either automatic or manual process is
performed by the judgement of operator. The automatic
detection realizes automatic detection of pass points with high
accuracy from stereo model (Sakamoto et al., 1998). The
principle is based on point matching with improved relaxation
method and a mathematical model. If the automatic process
fails to detect matching points, manual detection by operator is
performed.
In the next step, relative orientation parameters are calculated
with matching points detected above. If the calculation is
successful, formation of stereo model by image rectification is
followed. Different from normal stereo model of aerial images,
some case has been confirmed that the adequate stereo model
cannot be generated since the difference of the altitude of
photographing sites in two images. In this case formation of
stereo model by perspective projection is applied instead of
image rectification.
Image rectification is based on epipolar geometry, which
reduces the direction of mapping (matching) to x-direction only.
