In order to discriminate cloud and blue sky areas from the 
whole sky imageries, we use Sky Index (S7) and Brightness 
Index (BI) which are calculated from RGB channels of the 
image. SZ shows the blueness/greyscale and BJ shows the 
brightness in pixels. SI and BI are expressed by equation (1) and 
Q). 
DN; — DNp 
Sl = Br n (D 
DN5 + DNg 
DN, + DN , + DB 
Pr = 2 68 LR (2) 
@r—1) +3 
where DN, ; digital number of Blue channel 
DN ; digital number of Green channel 
DN; ; digital number of Red channel 
n ; bit number of quantization level 
SI values have the range between —1.0 and 1.0. The blue sky 
area in the RGB image has the higher digital number in blue 
channel and the lower digital number in red channel. On the 
other hand, the clouds show white or grey on the image. Thus, 
the higher value of SI shows bluer sky, the value of near zero 
shows the clouds and the sun. B/ values are expressed by the 
range between 0 and 1.0. The bright pixels on the image show 
high BI. The pixels in the sun area have B/=1 and SZ =0. 
The whole sky areas consisted of the sun, clouds and blue sky 
are classified by the threshold curve which was derived from 
two dimensions of SI and BI coordinates. The threshold curve 
for sky area classification was determined by counting the 
frequency of SZ and BJ combination’s pixel. In this case, the 
threshold curve was used as BI = e°”“, When the coordinates 
plot under the threshold curve, these are classified as cloud area. 
The coordinates upper the curve are classified as blue sky area. 
The sun area in the imagery has the coordinates of S7 = 0 and BI 
= 1. In this way, the sky classified image was generated by SJ & 
    
  
  
  
BI threshold. 
Whole Sky Image (t) 
| Sky Index (SI) | | Brightness Index (BI) | 
| Masking for out of area | | Masking for out of area | 
  
  
  
z. 
A it in. DS 
zz 
Classifying Sun, Cloud, and Blue Counting the frequency of SI 
sky areas by SI&BI threshold and BI combination’s pixel 
| 
Determination of SI&BI 
threshold for dividing sun, 
/ Sky classified image(t]/ cloud A areas 
Figure 1. The flowchart of cloud, blue sky and sun 
discrimination by whole sky imageries 
  
  
  
  
  
  
  
  
  
  
   
   
   
   
   
   
   
   
    
     
     
  
  
  
  
  
  
  
     
     
   
   
     
   
    
    
    
   
    
   
    
    
   
   
    
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
3. OVERLAYING TO SATELLITE IMAGE 
As for the method of overlaying with cloud areas of both 
ground and satellite-based images, the cloud area on whole sky 
image are projected to ground surface plane with map 
coordinate. Our used fisheye lens has equidistant projection. In 
case of setting the ADFC fixed to north direction and levelled 
horizontally, the central pixel of whole sky (hemisphere) image 
corresponds to the zenith. 
According to equidistant projection, the distance from the image 
centre is in proportion to zenith angle shown in equation (3). 
deg 9 
R ^ 90 e 
  
where — d,4; distance from zenith (image centre) 
R; Radius of whole sky image area (pixels) 
6 ; Zenith angle (degrees) 
Consequently, the zenith (0) and azimuth (d) angles on whole 
sky image are able to calculate from the image coordinates (u, 
v). And then, the horizontal distance (HD) from the ground- 
based observation point, that is central position of whole sky 
image, to the target cloud can be estimated by given the 
approximate cloud height supposed from each cloud type, and 
the cloud areas on whole sky image are projected to map 
coordinates (x, y) using 6, gand HD. 
In these calculation and conversion, any distortions of fisheye 
lens and the curvature of ground surface are left out of 
considerations. 
3.1 Calculation of zenith and azimuth angles 
Calculation of zenith and azimuth angles at any hemisphere 
image coordinates can be shown in equation (4) and (5). 
EN TTA CER = (4) 
when 0 € 0 « 180, 
  
  
pn = cos it o =) 
@; = 180 — cos ( R x 8; 
(5) 
when 180 < 0<360, 
Vi-—uv 90 
@; = 180 * cos! ( > : x 3) 
where  u;, v; ; Image coordinates of column and row (pixels) 
u,, v, ; Image centre coordinates (pixels) 
0. ; Zenith angle (degrees) at (uj vj) 
@; ; Azimuth angle (degrees) at (u;, v;) 
In this study, the image size of hemispherical area has 1650 
pixels and 1650 lines. Therefore the radius of circle (R) is 825 
pixels and the image centre coordinate (u,, v,) is (824, 824) in 
case of the image upper left coordinates as (0, 0). 
3.2 Cloud area conversion to map coordinates 
For each pixel classified as cloud area on whole sky image, the 
projected cloud area on the ground can be converted to map 
   
Intern 
coordinates 
any cloud I 
stratocumul 
of cumulus, 
The conver 
whole sky 
equation (6 
where 
OC MOM NO ~~ 
As one of ( 
several wl 
synchrono 
Kyoto, Ja 
located at 
we use UT 
Figure 2 si 
whole sky 
  
Figure 2. 
Cloud dis 
image. ]