BUILDING DAMAGE DETECTION FROM POST-EARTHQUAKE 
AERIAL IMAGES USING WATERSHED SEGMENTATION IN GOLCUK, TURKEY 
E. Sumer ?, M. Turker b 
? Baskent University, Department of Computer Engineering, Eskisehir Road 2 
0.km 06530 Ankara, Turkey — 
esumer@baskent.edu.tr 
^ Middle East Technical University, Graduate School of Natural and Applied Sciences, 
Geodetic and Geographic 
Information Technologies, 06531, Ankara, Turkey — mturker(cümetu.edu.tr 
Commission VII, WG VII/5 
KEY WORDS: Remote Sensing, Earthquakes, Detection, Segmentation, Aerial, Edge, Urba 
ABSTRACT: 
The collapsed buildings due to earthquake were detected fro 
objective was to detect the collapsed buildings based on the 
stored in a GIS as vector polygons. The building polygons were utilized to 
n area of Golcuk. The shadows cast by the buildings were detected using the watershed 
dges of the buildings were identified and a buffer zone was generated for each building 
within the buffer zone were selected from both inside and outside the 
shadow regions were detected using a watershed segmentation algorithm. 
adow producing edges of the buildings and the corresponding shadows 
pixels. Of the 284 buildings analyzed, 229 were correctly labeled as collapsed or un-collapsed 
The results prove that the collapsed buildings caused by the earthquake can be successfully 
approach was implemented in a selected urba 
segmentation algorithm. The shadow casting e 
polygon along these edges. Then, the initial points falling 
building polygons to start the watershed segmentation. The 
This was followed by measuring the agreement between the sh 
based on the percentage of the shadow 
providing an overall accuracy of 80%. 
detected from post-event aerial images. 
1. INTRODUCTION 
An earthquake, an unpredictable and unpreventable event, is 
regarded as one of the most destructive natural disasters on 
earth. On 17 August 1999, the urban areas of Golcuk, Yalova, 
[zmit and Istanbul were significantly damaged by an 
earthquake. It is estimated that 50,000 buildings were heavily 
damaged, over 15,000 people died and about 32,000 people 
were injured in this dreadful event. The epicenter of the 
earthquake was 40.70° N, 29.91° E (USGS), near the city of 
Izmit. The magnitude and the depth were 7.4 and 20 km 
respectively. The region struck by the earthquake, 
accommodates nearly 20% of the total population and is the 
most industrialized zone of Turkey. 
The extent of the damage caused by this catastrophic event 
needs to be assessed rapidly in order to reduce its effects by 
setting the corresponding agencies in motion. This can be 
efficiently performed using remote sensing technology that 
provides up-to-date information about the earth surface features. 
Change detection approaches can be used to detect the 
earthquake-induced changes using the pre- and post-quake aerial 
photographs or satellite images by comparing and analyzing 
them. There are also several other methods for collecting 
information on damage due to earthquakes such as field surveys, 
aerial television imagery, and satellite imagery (Hasegawa et al., 
1999). 
The objective of this study is to determine the collapsed 
buildings in a selected urban area of Golcuk using a watershed 
segmentation algorithm. A shadow-based damage detection 
method was proposed. The implementation of the approach was 
n, Building. 
m post-event aerial images using watershed segmentation algorithm. The 
analysis of the cast shadows. The building boundaries were available and 
perform assessments in a building specific manner. The 
carried out using MATLAB 6.5 which is a high-performance 
language for technical computing. 
2. PREVIOUS STUDIES 
In many applications of damage assessment and building 
detection, the aerial photography is widely used due to its 
advantages such as improved vantage point, permanent 
recording, broadened spectral sensitivity, the increased spatial 
resolution, and geometric fidelity. One of the frequently used 
applications of aerial photography is the detection of the 
buildings from their shadows. Irvin ef al., (1989) states that the 
shadows are usually among the darkest areas in images and their 
extraction can be feasible using image processing techniques. 
They developed several methods to estimate the grouping of 
related structures together with the shape, verification and 
height of individual structures. In each method, the main 
approach used was the relation between structures and their cast 
shadows. Another study concerning building shadows was 
realized by Huertas ef a/., (1988). They used building shadows 
to estimate the building heights. In addition, the shadows cast by 
the buildings were utilized in verification of the buildings. Their 
method was comprised of four steps including line and corner 
detection, labeling of the corners based on shadows, tracing of 
object boundaries, and finally the verification of hypotheses. 
Ishii et al, (2002) proposed a method, containing two cases, to 
detect the damaged areas from aerial photographs. In the first 
case, color and edge information were used to detect the 
damaged areas from a post-quake aerial photograph. Combining 
the color information with the edge information, the 
542 
Inte 
disc 
coul 
pho 
eart 
usin 
the 
area 
cons 
in « 
conc 
bull 
the 
The 
dam 
brigl 
dam 
carri 
An 
integ 
Gam 
phas 
via 
infra 
real 
detec 
the p 
Turk 
imag 
chan 
chani 
event 
founc 
In a r 
colla; 
earth 
the ac 
earth« 
differ 
buildi 
The p 
as 84 
shado 
earthc 
to ma 
corres 
specif 
detect 
In th 
detect 
segme 
Waters 
topogr 
as at 
image 
image 
Waters 
low br 
(Sonk: 
the mi 
relief i 
into th 
deeper 
to pre