Istanbul 2004
etector.
ameters were
er of pixels to
is required to
. The second
er of pixels to
rst and the N"
line by means
1e edge pixel
the last pixel
line segments
[3 pixels were
> vectorization
. To group the
od was used.
together to see
carried out to
n fragmented
| vectorization
! find a corner.
sessed.
s belonging to
| do that two
i) collinearity.
tween the line
tation between
was selected to
the buildings
segments that
usly combined.
3.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
sat Building
et G
DE 4 SP x
; nn ;
f fu à ^, f
4 f IL Jl 7 Le ;
s y ae z E i
: ie fes © / New Line
(jg Es v / ombine
5 2
uu UE 2 in /
137 i. f 2 Y
m $
3
d
Figure 3. The first level perceptual grouping.
3.2.2 Second Level Grouping
The study area contains usually the rectangular shaped
buildings. As can be known, the edges of a rectangular
shaped building intersect at the corners with an angle around
90°. Therefore, in the second level grouping, the corners were
used as an indication of a building and the line segments were
grouped according to the principles of the perpendicularity
and the proximity. For each couple of line segments, these
two principles were checked whether they were satisfied or
not. When a couple of line segments were detected then,
these two line segments were grouped together. The second
level grouping is illustrated in Figure 4.
,-- Line Segments
mmm à
t = B CI E. ;
ry À € a : ;
ETT ies S. ^77, Maximum Distance
E e i
| IL | = a m Angle Between Lines
E nib IE
| —————— 4
Figure 4. The second level perceptual grouping.
3.3 Assessing the Conditions of the Buildings
After grouping the line segments through the perceptual
grouping procedure, the conditions of the buildings were
assessed on a building-by-building basis. For each building,
the assessment was carried out based on the measurement of
the agreement between the above detected line segments and
the vector building boundaries. This is based on an
assumption that if the vector building boundaries match with
the detected line segments then, the building under
consideration is declared to be un-collapsed. To measure the
degree of the match between the line segments and the vector
building boundaries, three parameters were used: (i) the
orientation, (ii) the distance between the line segments and
the edges of the building polygons, and (iii) the length of the
line segments.
Of these parameters, the orientation was used to measure the
degree of parallelism between the detected line segments and
the edges of the vector building polygons. In the present case,
the value for the orientation was set to 10°. The distance
between the detected line segments and an edge of a vector
building polygon shows how close the line segments are to
the edge of the vector building polygon. The closer the line
segments to an edge of the building polygon the higher the
chance that they belong to that building. The third parameter
measures the degree of the coincidence between the line
segments and the edges of the building polygons. If a
building is collapsed then, the degree of the coincidence will
be low. On the other hand, the un-collapsed buildings are
expected to show a high degree of coincidence.
This is illustrated in figure 5 where, 11, 12, 13, 14, and 15
represent the line segments detected through perceptual
grouping procedure. The broken lines illustrate the
boundaries of the vector building polygon. As can be seen in
the figure, there is a full overlap between the line segment 11
and the left edge of the building polygon. The overlap
between 12, 13, and 14 and the upper edge of the building
polygon is about 75%. On the other hand, approximately
50% overlap is measured betwen 15 and the right edge of the
building polygon. In the present case the threshold value was
taken as 75%.
Overlapping parts
J hw
| N ;
12 ^ a à ai Line segments
1 £4 1 | 1 I 1
i : LU à et /
7
15/
/
eS m IZ he "uU
7 d
f
Building
Figure 5. The overlaps between the line segments and a
building polygon.
The final decision about the damage condition of a building
was made based on the degree of overlap between the edges
of the vector building polygon and the above detected line
segments. If at least three edges of the building polygon show
an overlap above the pre-set threshold of 75% with the line
segments then, the polygon is declared to be un-collapsed.
However, due to the illumination effect and the type of the
roof material, the contrast between the roofs and the
surroundings may not be high and therefore, those edges of
the buildings located in the opposite direction of the
illumination may not be detected by the Canny edge detector.
As a consequence, because of the misdetected edges, the un-
collapsed buildings can be labeled wrongly as collapsed. To
overcome this problem, the shadow producing edges of the
buildings were tested. If the shadow producing edges of a
building show an overlap of above the pre-set threshold value
with the line segments and if there is a shadow corner formed
by these edges then, the building is labeled un-collapsed.
4. RESULTS
The assessment results of the proposed damage detection
approach is given in table 1. The results show that the
proposed approach for detecting the collapsed buildings due