0 9
1
P
= ER I e Viu olm SAEI A ed ae a na et de
expressing radial symmetrical and decentered lens distortion in
addition to the calibrated focal length, the point of best symmetry
and, in calibration 6, the principal point of autôcollimation. All
other calibrations provide a setup with reference to the principal
point of autocollimation, the measurement of the distortion with
reference to this point and an adjustment to find a point of best
symmetry. '
Rather than comparing certain values, as was done in the last
section, one can also investigate the effectiveness of the correction
of measured image coordinates using all parameters of the interior
orientation, for example by means of photographs over a photogrammetic
test field. Such an investigation has been carried out at The Ohio
State University in the course of a doctoral program by Jiwalai [15]
under the advisorship of Prof. Dr. D.C. Merchant [16], the vice-
chairman of the Working Group on Image Geometry, using RMK photography
obtained for this working group over Rheidt Test Field near Bonn,
West Germany (BRD).
First, a single photograph with 34 uniformly distributed points
was used. All calibrations resulted in very similar residuals and a
2.test at the 5% level of significance did not provide evidence to
reject the null hypothesis that all results have the same mj.
Table 3 lists some of the results obtained, Table 4 a typical
correlation matrix which indicates strong correlation between x. and
and between Yo and ® . The principal point shifts resulting
from the rotations o and o are more accurate and almost always
smaller than the actual shifts. The largest rotations are obtained
for calibration 2a, the only one disregarding the tangential
distortion. The variations in Zo indicate differences in the
calibrated focal length; it was determined for the reseau plane
in calibrations 2a, 3a, 4b and 6, and for the image plane for
calibrations 5 and 7. It appears that the spacing between reseau
plane and image plane of approximately 25 um was not taken into
account properly. The values obtained by calibration 6 and listed
in Table 1 deviate significantly from the others.
Second, one of the resections was repeated, now using 34 groups
of three relatively closely spaced points instead of 34 points. The
obtained root mean. square positional errors within the groups
(m = * 3.3 um,m = * 2.9 um) indicate, when compared with the overall
rms positional errors (m = 4.3 um and mi 5.0 um), that some
uncompensated systematic error is present in the data.
-19-
dh
FU
Ly
1
L|
HN
a
E
E
Y
L3