International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Although visual interpretation is a powerful tool, it is a
subjective and time-consuming process, which is of critical
importance in the time of emergency. To overcome these
problems, digital analysis of video imagery was carried out.
The methodology was tested on six representative frames,
which were selected according to different damage types in
different parts of Golcuk city. Hue, edge, saturation variance
and edge variance were used as feature layers, and the pixels
values of each layer were allocated a l-byte value. The
intensities of edge elements were calculated by a unidirectional
Prewitt-gradient filter with a 3x3 window size. Local variance
of feature layers was analysed for the area of 5x5 pixels.
Threshold values of damaged areas were determined according
to mean and standard deviation of pixel values of training data
sets selected from the reference frames. +/- standard deviation
from the mean value was used for multilevel thresholding for
each frame (Hue: 38-126, Edge: 6-74, Saturation Variance: 76-
198, Edge variance: 0-25). After the multilevel thresholding
process, a 61x61 mean texture window, chosen based on the
average building size in the video frame, was used to aggregate
and remove spurious pixels. The results of digital analysis of
video imagery are shown in Figure 8. The results were
compared with actual damage information observed from video
imagery (Figure 9). Overall accuracy ranged between 68% and
86% (Producer accuracy: 46%-83%, user accuracy: 47%-73%).
| DAMAGE INFORMATION DERIVED FROM VIDEO IMAGERY |
LEGEND
Damage information
[| undamaged areas
Ed Damaged areas
Fieure 8. Damage information derived from video image
g g
Further investigation was carried out to determine the
recognizable and unrecognisable damage types using digital
analysis of video imagery (Figure 9). The result shows that
digital analysis of video imagery failed to detect intermediate
and first floor collapse, but was effective in detecting rubble.
On the other hand, scale variations between proximal and distal
parts of the frame created failure in damage detection, as some
distal areas have the same textural features as rubble. In
addition, some building façades, which are under construction,
were also identified as damaged.
In conclusion, the use of highly oblique aerial video imagery
improved damage assessment compared with Spot imagery, as
it allowed seeing the facade of the buildings. Although there are
significant difficulties in digital image processing, the accuracy
assessment results are promising and encouraging for further
research. Movement of the helicopter, an unstable camera and
frequent scale chances due to zooming resulted in poor quality,
blurred imagery, which creates difficulties in digital image
processing. Moreover, the highly oblique characteristic, scale
variation in and between frames, heterogeneous and large size
pattern characteristic of urban scene are other obstacles for
digital image analysis of video imagery. In addition, there are
also difficulties in mapping damaged areas derived from the
analysis, due to lack of external and internal camera parameters,
and accurate ground control points required for the
orthorectification process. Moreover, methodology used in the
research is still data specific, as threshold values are determined
from training data sets.
RESULT OF ANALYSIS OF VIDEO IMAGERY
COMPARISON WITH ACTUAL DAMAGE
M [ ]carecty Mentied damaged arear
: ; ES hiorec Hy Idenited damaged areas:
à (3 Urdamagal areas
ME vrierted danayed ae
damage
4. CONCLUSIONS AND DISCUSSIONS
In conclusion, post-earthquake damage assessment in the case
of the 1999 Kocaeli earthquake has shown that Spot imagery is
of limited use for post-earthquake damage assessment due to
external factors and technical limitations (vertical viewing
characteristic, spatial and temporal resolution). In addition,
change detection gives information about change values of the
pixels, but not about the nature of the damage. Spot imagery
can provide overall information about concentrated and highly
damaged areas. With the integration of vector data,
visualization and interpretability of the results improved.
Although video data pose substantial processing, registration
and integration challenges, facade viewing characteristics
contribute valuable information. The results of visual
interpretation and digital analysis of video imagery have shown
that it improves the damage assessment at the local level
compared with Spot imagery analysis.
Analysis of user information requirements in the case of Turkey
showed that damage information varies depending on the
governmental hierarchy and activities of the agencies. There is
a strong need for base data integration with remote sensing data.
Spot imagery can be useful in strategic decision-making at the
national level and can guide airborne data acquisition at the
local level. Aerial video imagery can be helpful in coordinating
emergency activities and directing ground teams. For an
effective flow of information between different emergency
agencies, there is a need for improvements not only in technical
infrastructure, but also in organizational structure. As a
proposal of the research, the establishment of a Disaster
Management Centre at the national level, which is in charge of
setting up a spatial database, downloading and processing
facilities for remote sensing data, and information network, can
coordinate information flow between different users.
The International Charter on Space and Major Disasters, a
unified system of space data acquisition and delivery, and
Bilsat, the first Turkish Earth observation satellite launched
October 2003, part of the Disaster Monitoring Constellation, are
promising improvements for rapid data gathering and future
690
Int
apr
acti
Ove
vid
wit
ides
on.
pat!
thet
ext
All
Koc
Jap:
Soc
Aya
Schi
Sen:
and
haza
(acc
Bru:
Geo
Effe
9IS/re
Chir
Higt
Indiz
http:
e200
EDM
Aug
Rese
Ham
Stanc
http:/
-04.h
Jense
Rem«
Lond
Kerle
as a’
140-1
Mitor
Detec
Using
http://
tml (a
Monte
Mana
Geogr
Rodar
Multi
Engin
