intly
>, to
1s to
rain
gital
Ss, 2)
with
ul to
ndex
Ising
loud
; and
Ising
> sky
ı the
j)uds,
"ries.
1dex,
ginal
uture
lobal
Vers
poral
f the
; and
S.
5000
The
1280
mage
PEG
files
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Fisheye lens: The model of used fisheye lens is Nikon Fisheye
Converter FC-E8. The angle of view area is about 183 degrees
covered to hemisphere. FC-E8 has Equidistant Projection that
means the centre of lens is correspond to zenith, and zenith
angles are in proportion to distances from the centre point of
zenith. One of the other projections for fisheye lens is
Orthographic Projection that shows sine of zenith angles are in
proportion to distance from the centre point of zenith. Figurel
illustrates Equidistant and Orthographic Projections of the
fisheye lens.
Zenith Zenith
à à ^ c.
Hemisphere Vin Hemisphere Qi
"0% CO
d
Horizon
d a
Horizon Horizon
(b) Orthographic Projection
a) Equidistant Projection
Figure 1. Projections of fisheye lens
Left: Equidistant Projection, 0/90 = d,/R
Right: Orthographic Projection, sin 0= d/R
As the areas of which the zenith angles are wider in hemisphere
image taken by equidistant projection, these image areas where
is especially near the horizontal line are shown as larger than
the same areas in case of taking by orthographic projection.
That means orthographic projection is useful for calculating the
ratio of area. However, there is very little availability of these
kinds of fisheye lens with orthographic projection. Therefore, it
is necessary to do the cosine correction of zenith angle for each
pixel in the hemisphere image taken by the equidistant
projection when cloud cover is calculated.
Remote control cord: This cord, which is Nikon Remote Cord
(MC-EUI), is used for taking pictures automatically at the same
interval. It is able to set two minutes interval as the shortest. We
set two minutes interval.
As the other accessory, ND filter with 0.9 optical density
(transparency about 13%) is put between camera lens and the
fisheye lens in order to decrease the strong sun light and defend
CCD saturation.
2.2 Camera setup
This camera system is installed at the top of tower with 45
meters heights. The scene of sky area is almost hemisphere.
Here, various meteorological observation systems are equipped
in addition to the camera system.
The head of digital camera is fixed to north direction and
levelled to horizon on tripod. As for the exposure control, we
have priority aperture and set the aperture as f2.8 because the
image contrast is better than the image taken by the aperture
more than f2.8 or shutter speed priority.
The image size and quality should be different from the purpose
of use. The image with the full size and the uncompressed
quality might be needed for the purpose of precision
827
photogrammetry, but our use of images is to understand cloud
distribution and estimate cloud cover in this study, therefore the
image size and quality are set as 1024 x 768 and the
compressed JPEG format respectively.
2.3 Hemispherical area on image
Given this camera system and setup described previously, the
hemispherical area shown in the whole image is determined.
Figure 2 shows the hemisphere area and image coordinates in
1024 pixels and 768 lines.
UL- (0, 0) 1024pixels
(213, 107
X n
CUERO
580pixels
580lines
768 lines
Figure 2. Hemispherical area and image coordinates
In this case, the radius of the circle filled with hemispherical
area is corresponded to 290 pixels. One pixel corresponds to
about 0.31 degrees of zenith angle.
3. METHODS OF CLOUDS DETECTION
As the pre-processing of hemisphere imageries, each image is
clipped with 580 pixels by 580 lines, and then the masking
image of sky area except for the others around horizon is
generated (Figure 3).
1024pixels
580pixels
768 lines
580lines
$
c n
Figure 3. Pre- processing of hemisphere image
3.1 Detection of clouds area
Gabriela et al. (2002) used the simple ratio of red / blue from
RGB image to extract feature in sky image. Here, we also use
two channels of red and blue from the visible RGB image in
order to separate the blue sky and cloud areas. The blue sky
area in the RGB image has the high digital number in blue
channel and the low digital number in red channel. On the other
hand, cloud areas show white or grey in the image that means
the digital numbers of red and blue are almost same or close.
Therefore, as one of the methods for separating to blue sky and
clouds area, we use equation (1) and generate the index
