. Istanbul 2004
h x axis is the
ange detection
tching method.
esult, 80 % of
alse alarms are
itching method
ent
; process
1=5)
ection result
mage)
ching method
howed the best
9.0 for contrast,
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
0.5 - 1.5 for angular second moment, 0.0 for entropy and 0.0 -
1.0 for mean by some supervised learning experiments.
Another experimental result for change detection between aerial
photos in 1998 and 2003 shows 80% of right change detection
can be achieved when false alarms are about 6% in 3D image
matching method (Fig.15). The result of Fig.14 is poor
compared to this result. This is probably caused by long
passage years of the image to be compared and the influence of
clouds that are visible in Fig.9 (a).
(b) Photo interpretation result
(green: not changed, red: changed)
(a) Target area of photo
interpretation
Figure 12. Photo interpretation result
Not changed (Nt - Nc)
Over detection (Nf)
Under detection (Nc - Nf)
Right detection (Nr)
Figure 13. Change detection by 3D image matching method
TITAN IE TT TT T T m1
i -]
68r 7
0.8
0.6
PF/PD
PD
04 1
0.4
rr dx hoa 1 1 |l ero ries
0 1:2 3 4 5160 7 8 9 MW (in) 0 0.2 04 0.6 0.8 |
PF
Shift of building's height
Figure 14. Change detection ability by 3D image matching
method (1995 - 2003)
croi T TU TT ST T T TET
0.61
PF / PD
PD
04r 1
a la ps d
eie d: 1 1 1
0 12-3541. 5 6.7 8.9 10m) 0 0.2 0.4 0.6 0.8
PF
Shift of building's height
Figure 15. Change detection ability by 3D image matching
method (1998 - 2003)
6. CONCLUSIONS
In this study, we proposed a change detection approach
objected for metropolis right after the disaster with automatic
image processing of aerial imageries. We introduced two
different types of approaches that are 2D image matching
method and 3D image matching method. The first method is
for the case when no orientation information is available. The
second approach aims at acquiring not only 2-dimensional
changes’ distribution but also quantitative 3-dimensional shifts
by matching between former digital terrain data and images
after the disaster.
Evaluation tests were performed "with actual aerial imageries
that were taken right after the earthquake and several years later.
To evaluate change detection ability, comparison of change
detection results with photo interpretation by human operator is
utilized. For quantitative estimation ROC chart was applied,
which plots the sequential probability of detection against the
probability of false alarm. As a result, 80 % of right change
detection has been achieved when false alarms are about 30 %
in 2D image matching method and 18 % in 3D image matching
method respectively.
References
Kosugi, Y., Tchimev, P., Fukunishi, M., Kakumoto, S. and
Doihara, T., 2000. An Adaptive Nonlinear Mapping Technique
for Extracting Geographical Changes. Proc. GIS2000, CD-
ROM, 4pages.
Kosugi, Y., Fukunishi, M., Sakamoto, M., Lu, W. and Doihara,
T., 2001a. Detection of Sheer Changes in Aerial Photo Images
Using An Adaptive Nonlinear Mapping. Proc. DMGIS 2001,
pp.145-146 (Bangkok 2001).
Kosugi, Y., Nojima, Y. and Uto, K., 2001b. Coincidence
Enhancement As A Brain-like Information Processing Principle
Applicable to The Adaptive Nonlinear Mapping of Images.
Proc. SCI 2001, Vol.X, pp.331-336 (Orlando 2001).
Sakamoto, M., Uchida, O. and Wang, P., 1998. Automatic High
Accuracy Tie Points Detection in Stereo Images Using
Relaxation Method and Gradient Method. Journal of the Japan
Society of Photogrammetry and Remote Sensing, Vol.37, No.5,
pp.35-46.
Sakamoto, M., Kosugi, Y., Nakamura, M., Fukunishi, M., and
Doihara, T., 2001, Change Detection of Geographical Features
with An Adaptive Nonlinear Stereo Mapping Process, Proc.
LUCC2001, Session 5, R8, CD-ROM, 7pages, Tokyo.
Vantree, H.L., 1968. Detection, Estimation, and Modulation
Theory / Part I. John Wiley and Sons.
Weszka, J. S., Dyer, C. R. and Rosenfeld A., 1976, A
comparative study of texture measures for terrain classification,
IEEE Trans. Syst, Man, Cybern., vol. SMC-6, no. 4, pp.269-
285.
Acknowledgements
This work was supported by the Special Project for Earthquake
Disaster Mitigation in Urban Area.
