International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
Synthetic Aperture approach, with quantitative 
assessment using a the Universal Quality Index 
(Wang and Bovik, 2002); 
(it) Damage detection in video sequences using 
empirically derived colour indices, edge 
characteristics and their variances; 
(111) Automatic extraction of GPS information and 
camera orientation to map flight path and camera 
[FOV; 
(iv) Mosaicing of videoframes to create a composite 
view of the disaster area to facilitate orientation. 
3. RESULTS 
3.1 Frame quality enhancement 
The video data acquired, in addition to their inherent 
comparatively low quality, suffered substantial degradation as a 
result of a series of conversion steps. The imagery was recorded 
digitally (720 columns), and transferred to S-VHS tapes (420). 
These files were later copied to VHS (240) and made available 
to us. We employed a Sony TRV125 D$ digital video camera 
for an analogue-digital conversion (back to 720x576 pixels). In 
total, we obtained 25.5 minutes of coverage recorded on 13 
May 2000. Clearly, some of the conversion steps were quite 
unnecessary, their avoidance likely leading to improved damage 
assessment results. 
AstroStack and a Synthetic Aperture approach 
In order to restore some of the lost information and reduce 
overall noise, an image stacking procedure was performed in 
AstroStack (www.innostack.com). From a range of adjacent 
frames a reference frame was chosen, with which the other 
frames were correlated. Every frames was then shifted in x and 
y, as well as rotated with respect to the reference frame, to 
maximise correlation. For this maximisation the Universal 
Quality Index (UQI, Wang and Bovik, 2002) was calculated for 
every frame. The UQI also uses a correlation coefficient, in 
addition to comparing luminance and contrast. The resulting 
aligned frame series was averaged into a new image with 
reduced noise. 
Gornyi and Latypov (2002) recently described an image 
enhancement procedure using a synthetic aperture (SA) 
approach, whereby subpixel-size features were resolved from 
digital images. The principle involves an image series of an 
object, which is tracked in sub-pixel increments. Provided that 
the object itself does not change in-between frames, upon 
proper alignment of the frames a resolution can be achieved that 
surpasses that of the recording sensor. Gorny and Latipov's 
work suggested that, given a number of frames and and 
subpixel scanning of the subject, a substantial resolution 
increase can be achieved. To verify their results, and explore 
the applicability to enhance video data, we first set up a 
controlled experiment. A picture was produced with lines 1 and 
3 pixels wide, with in-between spaces ranging form 1 to 5 
pixels (Figure 2). Ten individual images were created with 
incremental 1-pixel horizontal shifts, played on a computer 
monitor and recorded with a Sony TRV125 D8 digital video 
camera. The imagery was processed within AstroStack, and the 
expected increase in resolution found compared to the 
individual video frames. 
  
   
  
   
uum 
un 
| EN ERE 
  
  
  
  
  
582 
  
Figure 2. Illustration of synthetic aperture approach to increase 
the resolution of individual video frames (see text for 
description and discussion) 
A total of 40 video frames (Figure 2b) of the original line array 
(a) was first simply stacked (top part of [b]), leading to an 
unfocused image. The lower part of (b) shows the result of 
stacking after alignment. A point spread function (PSF) derived 
from the actual lines in the lower part (b) was used for the 
restoration. The result of the restoration based on (b) is shown 
in (c). A much clearer restoration is shown in (d), for which the 
stacked image of (b) was doubled in size before alignment. The 
results show that details can be extracted that are not resolved 
in the original video frames. 
The SA approach was then applied to the police video data. 
However, the increase in detail observed in the controlled line 
experiment was not found. The likely reason for this is the 
accumulated video quality loss resulting from the 
aforementioned conversions. Especially the VHS conversion 
has led to smeared out details and line instability. The 
conversion to digital also introduced jpg-like artefacts, leading 
to further noise and reduction in dynamic range. Furthermore, 
individual elements in the video data contrast much less than 
the lines in the theoretical experiment. Such a reduction in 
modulation, however, increases the space between features that 
can be resolved. 
Although with a direct transfer of the original digital data to the 
computer the need for such resolution enhancements decreases, 
we expect the SA approach to be useful with higher quality 
data. 
3.2 Automatic damage detection 
The actual image processing to detect damage was carried out 
in a flexible processing environment created by Innostack. The 
software works with processing blocks that can be connected as 
required. The graphical user interface (GUI; Figure 3) displays 
the input images or video, the processed equivalent, as well as 
command prompt and history list. The GUI is customisable to 
allow easier execution of pre-defined routines. Our goal is to 
provide a working environment where post-disaster video data 
can be processed, spatially registered, and displayed together 
with co-registered pre-event images or map data as required. 
Interné 
Monitor 
genium 
  
Iv Fit 
Comms 
info bi 
info bt 
set bh 
sethh 
setbb 
sethh 
setbb 
setbb 
setbb 
setvie 
start b 
ad 
Figui 
Similar 
identifi 
intensit 
These » 
video s 
limited 
vertical 
overvie 
  
Tabl 
From t 
training 
convers 
clarity. 
elimina 
combin: 
satisfac 
similar 
and K. 
process; 
the resc