/6
of
an
id,
SHADOW ANALYSIS IN ASSISTING DAMAGE DETECTION DUE TO EARTHQUAKES
FROM QUICKBIRD IMAGERY
T. T. Vu *', M. Matsuoka *, F. Yamazaki *^
* Earthquake Disaster Mitigation Research Center (EDM), National Institute for Earth Science and Disaster Prevention
(NIED), 1-5-2 Kaigandori, Wakinohama, Chuo-ku, Kobe 651-0073, Japan - (thuyvu, matsuoka)@edm.bosai.go.jp
? Department of Urban Environment Systems, Faculty of Engineering, Chiba University, 1-33 Yayoi-cho, Inage-ku,
Chiba 263-8522, Japan - yamazaki@tu.chiba-u.ac.jp
Commission VII, WG IU/5
KEY WORDS: Earthquakes, Urban area, Change detection, High resolution satellite, QuickBird
ABSTRACT:
This study is to take the advantage of shadow appearance in Standard Imagery produced from QuickBird for damage detection in
urban areas. In a very complex scene of an urban area acquired from very high resolution satellite-based optical sensors, fortunately,
the buildings tend to align in some dominant directions in a small area and posses geometric regularity. Therefore, their shadows
also align following these dominant directions in a small area in spite of the acquisition condition. The changes of building
structures caused by an earthquake could affect the orientation, shape and size of its shadow. Two QuickBird scenes acquired over
the city of Boumerdes, which was one of the most heavily-damaged areas due to the Algeria earthquake of magnitude 6.8 on May
21, 2003, are employed in this study. The first one was about one year before the event (April 22, 2002) and the second one was two
days after the event (May 23, 2003). The result shows that the differences in shadow's lengths paralleling dominant directions can
assist the detection of collapsed buildings. Moreover, unlike other classes of land cover in urban areas, the shadows can be
successfully segmented by a conventional pixel-based classification method. The promising results from this analysis prove that
shadow-based information could be used as a potential cue for automated detection of building damage.
1. INTRODUCTION
Damage detection plays an important role in disaster
mitigation. It grasps the real situation after the events and
provides the required information for further damage
assessment. Hence, acquisition time, processing time, and
accuracy of detection are vital. Remote sensing techniques
providing the information in wide-coverage at a reasonable
time gap after the events has been increasingly considered and
employed. Recent researches in damage detection employing
remote sensing can be listed into two directions. While the first
direction is fusion of all available data sources as Casciati et al
(1997) fused GIS, space- and airborne imagery, the second one
develops methods for a specific kind of sensor. Typical
examples of the latter are Hasegawa et al (1999) with aerial
HDTV images, Matsuoka and Yamazaki (1999) with space-
based optical and radar data. In the context of quick damage
detection and consequently quick damage assessment, first
trend seems not to be always feasible. The reasons are unsolved
fusion methods and unavailability of GIS data in developing
countries. Following the second direction, we are focusing in
developing method for damage detection from very high
resolution satellite imagery such as QuickBird. Furthermore,
the trend of damage detection to automatic processing is also
our final goal.
Providing very high spatial resolution imageries, QuickBird
could provide enough detailed information for urban mapping
and hence, it also presents more complex scene of urban areas.
In such a scene, spectral information of building feature, which
*
Corresponding author
607
is our focus of interest, becomes diverse and ill-defined.
Conventional pixel-based image classification such as
parallelepiped, maximum likelihood, K-mean cannot produce a
reliable detected result in processing multi-spectral high
resolution satellite-based imagery (David and Wang, 2002).
Object-based methods seem to be promising approaches but
those methods have not been fully developed and implemented.
Fortunately, at least pixel-based classification can successfully
extract shadows. The stand of building casts the shadow in
surrounding. In other words, shadows convey some aspect of
information about buildings. The changes of buildings will
affect the shadows of the buildings and it is probably the cue to
assist the damage detection. Employing the well-developed
pixel-based image classification and investigating the
appearance of shadows to detect the damaged buildings are the
focus of this study. The following section presents some
observations which are the initial points for the shadow
analysis. It is followed by the proposed methodology for
shadow analysis, and testing result.
2. OBSERVATIONS
As abovementioned, Quickbird image presents very complex
scene of urban area. However, behind this complexity, within a
focused area, buildings tend to be aligned following some
specific direction (Sohn and Downman, 2002). It was
determined at the urban planning stage. Moreover, buildings
usually possess geometric regularity such as rectangle.
Therefore, building’s shadows will align following longitudinal
