ns were taken,
era in upright
ted 909 around
f 10 additional
, in order to
| introduced by
f the imaging
ated using the
(Beyer, et. al.,
neters entering
estimated as
crometers.
'ed the bundle
lution (only 7
ımed fixed). In
ct coordinates
1atic errors are
ibration model
| values for the
justments the
compensation
wn values only
the minimum
1 definition are
ites are known
rst facade, the
points and the
patible to the
in this CIPA
btained so far
frame with the
(Patias, et. al.,
ninimum norm
tment.
rained solution
ert transforma-
s) during this
] by the other
ctive variance-
to their free-
the adjusted
iteria, ranging
lual points) to
solution). It
curacy criteria
rom surveying
measurements) object coordinates of the witheld from the ^ adjustment check points.
Dataset | Adınstment | oo (um) | oy (mm) | oy (mm) | oz (mm) | vx (um) | v; (um)
I a 2.65 14.23 26.02 6.16 1.33 1.89
] b 2.58 8.32 4.67 9.18 1.46 1.65
I a 8.37 42.43 78.29 16.78 5.01 5.30
II b 6.70 25.52 38.52 13.39 6.89 5.94
I measuring device of 8.5 um a pre-calibration
II measuring device of 25 um b self-calibration
Oo a-posteriori standard deviation of unit weight
^ ^ ^
Ox Oy, 0, theoretical precision of check points coordinates in object space
X y
5.3.1 Local Criteria
Local criteria are the Mean Square Error o; of a point i,
the volume of the error ellipsoid, and the characteristics
of the error ellipsoid (lengths of the axes, etc).
2 2
Ox +0,+0,
3
Q =
1
5.3.2 Global Criteria
Although the use of the local criteria is useful for each
individual solution, for checking the accuracies in
different areas of the object, they are of little help in
drawing overall conclusions. Therefore additional global
type of criteria have to be defined and computed for each
solution.
Such global criteria can be the RMS value corresponding
to the mean variance D,, the Mean Standard Deviation
D;, and the Maximum Standard Deviation D,,,,.
$333 Graphical presentations
In order to graphically present the results, the following
plots have been prepared for each solution:
* Projections of the error ellipsoid of every point onto
the X-Y, Y-Z, Z-X planes.
Discrepancy vector plot, showing the differences
between the the point coordinates obtained through
the prior adjustemnts and those obtained after the S-
Vy » Vy rms values of image coordinates residuals
transform. These discrepancies can be used to
determine possible misfit between the individual
dataset and the common reference frame used for the
comparisons.
6. RESULTS
The results of the free-network sdjustment are shown
below. Plots of the discrepancy vectors and the
projections of the error ellipsoids are also shown.
As shown by the discrepancy vectors, the transformation
has been with no problems, which means that the two data
sets (the AUT and the ETH ones) are compatible and no
blunders exist. With respect to accuracy the more
conservative criterion D, showed a good agreement with
the more optimistic criterion of Dj.
Pre-calibration give rather good results of 1.5cm in object
space corresponding to about 6.5um or 76% of the pixel
size in image space, when the original images of 8.5um
pixel size is used. When the pixel size was blown up to
25um the results of course are getting worse and reach the
level of 4.5cm in object space or 19.5um in image space.
It is obvious that the obtained results are directly related
to the pixel size. That is 3-times bigger pixel will give 3-
times worse results.
Self-calibration improves the results by 30-40% relatively
to pre-calibration. Accuracies are getting better than 1cm
in object space corresponding to 4um or abour 50% of
the pixel size in image space, when the original images of
8.5um pixel size is used. When the pixel size was blown
up to 25um the results of course are getting worse but still
better than 3cm in object space or about 1/2 a pixel
(13um) in image space.
The obtained results are compatible with those from the
other participants. These participants have used regular
photography. The best results obtained by them are of the
order of lcm (highly accurate measuring device + self
461
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
