ul 2004
convert
are was
frames
feature
nd edge
used to
training
ti level
ysis, the
rmation
f aerial
hown in
yrmation
chy and
, there is
the local
the user.
ypes of
based on
15, only
ibitants
ency aid
isaster is
tion and
wing the
exposed
| without
iselves to
line data
ion, road
al for an
Aoreover,
ved from
nizational
allenge is
1 requires
ncies, as
the right
aluated at
nal level,
significant
is. Due to
there was
estimation
need for
ces in the
]uirement,
detection
ere other
between
ictors is a
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
challenging task without vector data integration. The results of
the change detection analysis were compared with damage
assessment information derived from ground survey (General
Directorate of Disaster Affairs). The comparison showed that
damage in the villages was not recognizable from Spot imagery.
However, positive change values of pixels indicating damaged
areas at the municipal and district levels were observed.
DAMAGED AREAS DERIVED FROM
THE CHANGE DETECTION ANALYSIS
LEGEND
Value
Will Minus change
—JNo change
ESI Positive change
Value
4 High : 254
Low: 36
Figure 4. Result of change detection analysis
At the local level, the analysis results were compared with data
from a ground survey carried out by the Architecture Institute
of Japan (AIJ). Damage information derived from the change
detection analysis was aggregated into parcel level to be
comparable. Integration with vector data improved visualization
and interpretability of the results, a critical requirement for the
user (Figure 5).
DAMAGED PARCELS .
DAMA GE INFORMA TION
DERIVED FROM SATELLITE IMA GERY
T Ta
LEGEND 5 LEGEND +
> Damaged parcels i
ws Heghly darnageed
mm Moderately damaged
was Slightly dams ged
CAected parcis
4 od an » ado CSS Eechiiteud parcels rene de |
Value Damaged areas $i
High : 255 BjPositive values
Low: 0
Figure 5. Comparison of original result of change detection
with vector data integration
According to the comparison, the highest correlation (0.205,
even thought it is still low) was found between damage level 5
(total destruction, AIJ ground survey) and totally damaged
buildings derived from Spot Imagery analysis. According to
Figure 6, Spot imagery failed to detect damaged areas in the
northwestern part of Golcuk city. Moreover, there was
overestimation of damaged areas in the central part of the city.
In conclusion, Spot imagery has significant limitations due to
external factors. Furthermore, change detection gives
information about the change in pixel intensity values, but not
about the nature of the damage, which is important for the user.
Despite its limitations, Spot imagery can be helpful to get
overall information about concentrated and highly damaged
areas,
689
COMPARISON OF CHANGE DETECTION (CD) RESULTS
WITH AIJ FIELD SURVEY RESULTS
dz
— X
}
LEGEND
EH Damage level 5 (AlJ)
(Damage level4 4 (AIJ)
C3Highly damaged (CD)
% (ZIModerately damaged (CD
1 | (1 Slightly damaged (CD)
[C_JParcel boundaries
8 u$ 25 300 36 1000
Figure 6. Comparison of change detection results with AIJ field
survey results
3.3 Analysis of Video Imagery
To improve damage assessment at the local level, oblique aerial
video imagery, which allows imaging of building façades, was
used in damage assessment. The first step in the analysis of
video imagery was visual interpretation for part of the area, as
explained below. Based on the oblique imagery, affected
buildings were classified as heavily damaged and totally
collapsed (Figure 7). Structural damages, indeterminable from
the building façade, could not be classified using video
imagery. The results of the visual interpretation were compared
with the Spot analysis results. The comparison shows that there
was significant improvement in the damage assessment at the
local level. Damaged areas in the northwestern part of the city,
which were not recognized in Spot imagery, were clearly
observed in video imagery. Moreover, more than 50% of totally
collapsed building, observed from video imagery, were detected
as non-damaged areas in Spot. This result also underlines the
limitations of Spot imagery in damage assessment studies.
However, use of video imagery taken by a media agency
created some limitations for the applications. First of all, as a
media agency collects information for the news, there was a
focus on highly damaged areas, making comprehensive damage
assessment impossible. Moreover, lack of coordinate
information and camera parameters created limitations for
locating video frames on the map. Therefore, in visual
interpretation, prior knowledge was used to locate damaged
areas.
DAMAGED PARCELS DERIVED FROM
VIDEO IMAGERY BY VISUAL INTERPRETATION
LEGEND
Dam age information
Will Collapsed
WB Highly damaged
EL iUndamaged
Parcel boundaries
M trs
# ns 7e Sid 786 1,000
Figure 7. Damage information derived from visual
interpretation of video imagery
i
y
