International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
b Damage assessment GUI 
    
-18{ x} 
  
Monitor 1 
   
  
  
  
  
  
  
  
  
  
  
  
  
  
se 
Iv Fit fv Aspectratio 
Command Center Result List = 
a info bbb ^l [ DEFAULT: bb4(variance): «PAR: ValueShiftz7; RegionSize=5; THmin=0.001; THmax-1.0; IN: INPUT1; C «| 
e y info bb7 | DEFAULT: bb5(pixelmap): «PAR: ShowSelectedzfalse; Rz0.8; G=0.4; B=0.2, RoiX=8; Roiy=70; RoiW=62 
an set bb1filename to P\Documentatie(TCprojectImages\TC_enschede-mitomi2UC_D_4imaNoenhRGB. ( DEFAULT-:bb6(damage) «PAR: Min-0 0; Max-0.5; RaiX-8, RoiYz70; RoiW-625; RoiH-435; IN: BASE; 
of set hb2.hmin to 0.001 | DEFAULT::bb7 (Save): «PAR: FileNamezimg/saved1.jpg; IN: Image; OUT: >} 
set bb2.hmax to 0.15 // select red, vellow [Parameter DEFAULT:bb1 filename setto «PADocumentatielTCprojectlmagesuTC, enschede-mitomi2! 
ed " eminta 2 [Parameter DEFAULT::bb2.hmin set to <0.001> ] 
set bh2.smin to 0 2 
he set bb2.smax to 0.5 [Parameter DEFAULT::bb2.hmax setto «0.15» ] 
; set bb2.imin to 0.7 [ Parameter DEFAULT.:bb2.smin set to <0.2> ] 
Wil set bb? max to 0.995 {Parameter DEFAULT::bb2.smax set to <0.5> ] 
he [Parameter DEFAULT::bb2.imin set to <0.7> ] 
he setview 1 to bb3.out 0 [ Parameter DEFAULT::bb2 imax set to <0.995> ] 
start bb1 = [DEFAULT::bb1 STARTED] 
ed « | ; [null] 
= [DEFAULT bb1 STARTED | - 
{start bb1 | | «| psg n 
ta. 
ne Figure 3. Screen capture of the processing environment for the video processing, with input video on the left, and damaged areas 
he coded in orange on the right 
he 
en Similar to the approach taken by Mitomi et al. (2000), we because of damage patterns (mostly rubble piles) that were 
ac identified representative damage areas and calculated hue, easier to distinguish from non-damaged areas. Applied to the 
y . . . . . . . . 
ng intensity and saturation, as well as edge and variance values. Enschede case data and using low-resolution data, considering 
re, These values were then applied on a frame-by-frame basis to a also unidentified damage and incorrectly identified damage, 
an video stream within the developed working environment. We only accuracies of 40% were possible using the parameters 
n limited damage assessment to the area within the horizontal and described above. 
nat vertical information bars (625x435 pixels). Table 1 gives an 
overview of the value ranges used. 3.3 Extraction of spatial information 
he Parameter Threshold range (0.0 — 1.0) Spatial data are only useful of they can be spatially referenced. 
es, A particular challenge of video data is their erratic nature 
ity Hue = caused by frequent cuts, and focal length (zooming) and 
ue ).0 ; 
Saturation 02-05 inclination changes. Use of the resulting data, therefore, 
Intensity 0.7 — 0.995 requires extensive local knowledge. The camera used by the 
Edge - Dutch National Police recorded horizontal GPS information, as 
dge (3x3) 0.0 —02 
Variance (7x7) 0.016 - 0.11 well as data, time, and relative camera azimuth and inclination 
A (see Figures 3 and 5 (d)). No data were encoded on flying 
e height and camera zoom factor. 
as Table 1. Threshold values used for the damage assessment 
ys A routine was implemented in the processing environment to 
as From the calculated means and standards deviations of the extract the available information. Given that the position of the 
to training areas it became clear that colour bleeding and format- information in the video frame is fixed, three areas of interest 
to conversions had reduced overall image contrast as well as edge were defined, for the date/time/GPS information, and the 
ata clarity. In particular also the conversion to low-resolution VHS azimuth and inclination arrows. A total of 12 correlation blocks 
rer eliminated significant detail We found that no threshold 
combination of different parameters was able to detect damage 
satisfactorily. This approach, however, performed well in a 
similar study of the 1999 Kocaeli (Turkey) earthquake (Ozisik 
and Kerle, 2004), which was also implemented in the 
processing environment presented here. This is likely because 
the resolution of the input data was higher (720x576), but also 
583 
was then defined (for numbers 0-9, and for the arrows). Given 
the large number of frames (15 per second), only every 15“ 
frame was processed, to obtain one measurement per second. 
The azimuth and inclination values range from —180 to 180, and 
15 to —120, respectively. The x, y location of the arrows in the 
processed frames was converted to azimuth and inclination by 
linear interpolation. The individual correlation blocks were