2. STUDY AREA AND DATA
The selected study area is the part of the city of Golcuk,
which is one of the areas most strongly hit by the earthquake.
It is located on south coast of Izmit Bay, which is east-west
elongated structural basin situated along the North Anatolian
Fault (NAF) at the eastern margin of the sea of Marmara. The
study area contains a total of 282 buildings. Of these
buildings, 79 were fully damaged and collapsed and the
remaining 203 buildings were un-collapsed. The post-event
aerial imagery dated September 1999 (1:16,000-scale) was
used for the analysis (Figure 1). The imagery was supplied by
the General Command of Mapping (GCM) of Turkey.
Figure 1. Study Area.
3. THE METHODOLOGY
3.1 Edge Detection and Vectorization
First, the Canny edge detection operator was applied to the
post-event aerial imagery in order to extract the edge pixels
between the buildings and their surroundings. The reason for
choosing the Canny edge detector was its efficiency and the
output it provides as one pixel wide edges. The one-pixel
wide edges are used in turn as the input for the vectorization
process. To apply the Canny edge detector algorithm, a built-
in function of Matlab 6.1.0 was used. The output of the
Canny edge detector is given in Figure 2.
Then, the output edge image was converted into vector line
segments using a raster-to-vector conversion process. During
the vectorization process, the locations of the vertexes on the
edge pixel segment were found. Two vertexes represent a line
segment, which may be a candidate for an edge of a building.
In other words, each vertex defines a terminal point of a line
segment. Therefore, it is important to find the locations of the
vertexes.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Figure 2. The output of the Canny edge detector.
To run the vectorization algorithm, two parameters were
used: (i) tolerance value (g) and (ii) the number of pixels to
recognize a vertex (N). The first parameter is required to
decide if an edge pixel is a vertex or not. The second
parameter is used to define the minimum number of pixels to
search a vertex. A line is drawn between the first and the N^
pixels. Then, a pixel that violates the straight line by means
of the tolerance value is searched for on the edge pixel
segments. This process continues from first to the last pixel
of the edge pixel segments and therefore, the line segments
are generated. In the present case, 1.5 m and 13 pixels were
used for e and N respectively.
3.2 Perceptual Grouping
The line segments generated above through the vectorization
process were available for perceptual algorithm. To group the
line segments, a two-level hierarchical method was used.
First, the colinear line segments were grouped together to see
if they are closely located. This process was carried out to
construct a full line, which might have been fragmented
somehow during the edge detection and the vectorization
steps. Then, the lines were grouped together to find a corner.
Finally, the conditions of the buildings were assessed.
3.2.1 First Level Grouping
In the first level grouping, those line segments belonging to
an edge of a building were combined. To do that two
parameters were used: (i) proximity and (ii) collinearity.
While the proximity refers to the distance between the line
segments, the collinearity measures the orientation between
them. In the present case the proximity value was selected to
be less than the minimum distance between the buildings
present in the study area. Otherwise, two line segments that
belong to different buildings would be erroneously combined.
The first level grouping is illustrated in Figure 3.
Internati
3.2.2 Se
The stu
building
shaped |
90°. The
used as
grouped
and the
two prit
not. WI
these tw
level gre
3.3 Ass
After g
groupin
assessed
the asse
the agre
the ve
assump
the de
conside
degree
building
orientat
the edg
line seg
Of thes
degree
the edg
the val
betwee
buildin,
