| 2004
n of
can
can
ing
s, a
1ed,
and
ony
[er-
cu-
ali-
ks
ge
is:
cel
cel
nd
ve
on
nt
1S
E parameter A,
0.147 ‘| My parameters A, B, B,
parameters A, A», Ag B,. B,, C
5] Chebycheff polynomials
0.4 --| [7] no distortion parameters
0.04
0.02
0.005
Maximal normalized span of principal distance
©
©
©
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
Figure 3: Maximum normalized range of changes in principal distance c. The spans are normalized with the image width.
From left to right: Logitech camera(1), Terratec camera (2), HP camera with VGA and 3MB resolution (3), Sony camera
with vga and SMB resolution with 4 different zoom values (4), Kodak DCS 460 without demounting of lens (5S), with
demounting of lens (SU).
ratio in principle depends on the used calibration model.
Therefore we give the values for all five calibration models.
We find the following:
1. For the Sony camera (4) we investigated four zoom fac-
tors. The largest variation can be found for a zoom factor
between 2x and 4x: Here the control of the lens is not be
performed accurately. Choosing zoom factor 1x leads to a
stable principal distance, because it's the limit of the zoom.
The zoom factor 4x also is more stable.
2. The highest stable in principal distance shows the Kodak
DCS 460 (5), as to be expected. No differences could be
found between the situation when leaving the lens mounted
or unmounted lens between two calibrations. The varia-
tions are less than 3 pixels, thus appear to be much smaller
than those reported in (Maas and Niederóst, 1997).
3. The HP (3) camera also shows a good stability in prin-
cipal distance because of its fixed lens.
4. The Terratec web cam (2) shows significantly higher
variations in principal distance than Logitech's web cam
(1). Reason: May be the large compression ratio of 15.9,
compared to the low ratio 2.9 of the Logitech web cam.
Observe, the variations in principal distance do not vary
for different calibration models, except for the case where
no compensation for non-linear distortion is provided.
3.2.2 Temporal changes in principal point. The tem-
poral changes of the principal point are shown in figure 4.
Again we relate the changes to the image width, and give:
max y Gn. — zg)? * (ym, - yu;)^/w
We find the following:
Un
o
1. The temporal changes in principal point show a depen-
dency on the used camera model, this dependency is simi-
lar for all cameras.
2. Again, the Kodak camera (5) is the best. But it shows a
significantly less constant principal point when unmount-
ing the lens between calibrations.
3. The HP camera (3) has the same stability as the Kodak
camera.
4. When changing the zoom of the Sony camera (4), the
principal point is more stable when using the full resolu-
tion.
5. Calibration without distortion parameters results in large
differences of principal distance and principal point be-
cause of improper modeling. However, this effect does not
occur when using the HP camera (3), because this camera
has very low distortion.
3.2.3 Temporal changes in distortion. Changes in dis-
tortion are given in figure 5, again referring to the image
width.
We find the following:
]. The largest changes in distortion are observed for the
Terratec web cam (2). Due to high compression ratio non
of the calibration parameters could be estimated as accu-
rate as for the other cameras.
2. Largest instabilities occur when using all distortion pa-
rameters. This modeling results in no clear minimum in
the bundle adjustment, and can be explained by the inclu-
sion of the radial distortion parameter A3, which cannot be
determined stable, cf. (Burner, 1995).
3. The stability of distortion estimation with the Kodak
camera is not significantly better than HP, probably, be-
cause the Kodak camera has significantly larger distortion
values.
