International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
2.2 Target field and camera configurations
A 2-D target field is made of 500 x 500 x 5mm metal plate, 24
x 33 (792 in total) retro-targets with the diameter of 3mm are
placed on a lattice of 15mm width.
A
Figure 1. An image of close exposure over the target field
The camera used was Kodak DCS660m (Monochrome, 2008 x
3040 pixels CCD) with Nikkor 20mm lens. In order to stabilize
the interior orientation parameters, working parts of the lens
was fixed with silicone. In exposures shutter speed was set to
1/400, aperture to F/22, and a strobe light was used.
The following three basic camera configurations were
considered.
(1) Panoramic convergent (Pan ) configuration
(2) Close convergent (Close) configuration
(3) Close parallel or Block (Block) configuration
(1) Pan: Two images were taken at every eight camera stations
with a convergent angle of 30degrees. The camera was rotated
by 90degrees at each exposure. The distance to the field from
the camera was 2,000mm. The object space was imaged to
about 600 x 600 pixels in the CCD area. This configuration is
expected to yield the high precisions of object space
coordinates.
(2) Close: Sixteen images were exposed at eight camera
stations just as in the same conditions as in (1), except for the
distance from the camera to the field, which is set to 500mm.
With this configuration a strong network is formed and the
entire sensor area is uniformly covered with target images.
(3) Block: At the six camera stations of 1,000mm from the
target field, two images per station were taken in the parallel
camera configuration. Each two images covered a 1/6 target
field as shown in Figure3 and were rotated with respect to the
camera axis by 180degrees to the other.
This configuration is based on the following idea. For
strengthening a network, convergent exposure is desirable.
However, convergent exposure deforms a target to an ellipse
form. And the close-up camera configuration may shift the
image centre of the ellipse from the true centre of the target,
which deteriorates image coordinates.
On the other hand, in a parallel exposure configuration, in spite
of a weak network, target images become homogeneous. That is,
this configuration has an advantage of the homogeneity to
increase in coordinate quality. Taking it into consideration that
the space coordinates of targets arc already determined, it is
expected that the weak network pose relatively little influence
on coordinate quality.
In order to keep target images homogeneous, it was necessary
to avoid extreme close exposures and cover the target field by
muli-exposures.
Figure 2. Panoramic and convergent exposure
Figure3. Block exposure
Including these three camera configurations, their combinations
were tested on calibration quality.
(1) Adjustment of only Pan images
(2) Adjustment of only Close images
(3) Simultaneous adjustment of Pan and Close images
(4) Simultaneous adjustment of Pan and Block exposure images
(5) Simultaneous adjustment of all the images
2.3 Coordinates measurement and adjustment
Since the target field is of a form of a lattice, it is easy to
identify image points using the 2D projective transformation
equation once the image points are extracted. Namely they can
be identified by manual labelling of four or more points of the
corners of the lattice. Hence the difficulty lies in measurement
of target images coordinates.
For this measurement, target images are first recognized by
binarization of images, and then coordinates are measured by
simple centroid calculation in Pan and Block images, while this
technique is hard to adapt to Close images, since the size and
brightness of target images vary drastically over an image plane.
The farthest target image is dark and its diameter is about only
three pixels, while the nearest target image is very bright with a
diameter of about 20 pixels. For this reason, simple binalization
does not work well. To conquer this difficulty, the Laplacian of
Gaussian filter in equation 2 was applied to the image, and the
target images were extracted using zero-crossing information.
Interna
JG
where (
An exa
part we
(left sic
When |
target i
a targe
circum:
image «
Accord
the diat
for the
5-20 pi
for p=
Tablel
deviatic
Becaus
with lit
exposui
3.
The ac
five cal
Figures
scale b
have be
For Sc:
[5mm
For the
applied
All th
(Univer
accurac
Three
camera
distanc«
converg
bundle-
parame
the resu
Then tt
and 12.
values 1
The siz
the targ
result, v
in Tabk
