ght (S) for
ation at ETH
ts. Cit Pts.
O © 0 >
E
OM oó
= WN
He
NO Nu
ght':(s,) for
on Melbourne
S. Ck Pts.
i3.
11.
10.
WNW
CS OO. n
13.
10.
10.
E UI Ul *o
=
e
WNW
accuracy of
, number of
1€ Lagrange
The results
nevertheless
se aspects.
mials
iomials have
rpolation of
rameters for
1992; Kornus
arse of the
S decided to
Lp. orders for
ransformation
second- and
nt be most
S of the
rs suggested
1 was locally
linear. Moreover, there was initially
some suspicion that bundle adjustments
with a limited number of ground control
points and third-order interpolation
functions were subject to a measure of
ill-conditioning which gave rise to
possible projective compensation
problems. This suspicion was reinforced
by the results vofr/a number'sof!.^free-
network‘ adjustments that were carried
out (see Fraser & Shao, 1996).
The results listed in Tables 2 and:3
support the use of second- or third-order
functions for the exterior orientation
elements. It is noteworthy, however, that
in the case of only four ground control
points, ‘the combination of third-order
functions for positional parameters and a
first-order model for attitude parameters
yields the. most .accurate. solution for
both data; sets, irrespective. of the
number of OIs.
5.3 Number of Orientation Images
The choice of the number of OIs is
dependent upon a range of factors. These
include the ability of the Lagrange
polynomials of a given order to
adequately model the temporal variations
in position and attitude of the sensor
over the distance between adjacent OIs.
They also include consideration of the
number and distribution of available
ground control, which can impact upon the
stability of the resulting normal
equation system of the bundle adjustment.
For this investigation, computations were
performed with four, Six and eight OIs,
which corresponds approximately to an OI
every 83000, 5300 ‚and 4000 CCD lines,
respectively. The time interval for the
polynomial approximation in the case of
eight OIs is about 8 sec.
As can be seen from Tables 2 and 3, there
is little distinction between the results
obtained for the cases of six and eight
OIrs, at least for adjustments
incorporating polynomials of higher than
first-order. This is also the case with
changing the control point distribution,
but here the two image data sets show
different trends. In the 2-ray
triangulation adjustment with 4 control
points there is an increase in accuracy
when adopting eight OIs instead of six.
With the triangulation of the three-fold
stereo imagery the opposite occurs,
though in neither case are the changes
significant.
Bundle adjustments with four Ols were
also carried | oul. The triangulation
accuracies obtained were significantly
worse in height when first-order
interpolation functions were employed,
but were otherwise only marginally
inferior, by 1-2m or so, to the
corresponding accuracies for six and
eight OIs.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
5.4 Control Point Distribution
The five adopted ground control point
configurations are indicated in Fig. 1.
It should be recalled that the accuracies
listed in Tables 2 and 3 represent the
mean values obtained from adjustments for
Sets 2 and 3 for the case of 12 control
points, and for Sets 4 and 5 for the case
of 20.
It is apparent from Table 2. that;;the
number of control points has very little
impact on the accuracy of ground point
determination. It is hard to say whether
this is a reflection of a strong
‘relative orientation’ of the three-fold
stereo imagery or due to accuracy trends
being concealed through the influence of
residual systematic image errors coupled
with control point identification
problems. A similar masking of trends
might well be at work in Table 3 where
the results generally show an improvement
in planimetric accuracy with the
provision of additional control points.
With heighting accuracy the situation is
reversed, but again, the variations in
accuracy are by no means significant.
6. CONCLUDING REMARKS
From the results obtained from
triangulation of the imagery covering the
Australian Testfield, it can be safely
concluded that ground point determination
to 10m (0.7 pixel) accuracy in planimetry
and close to 6m (0.5 pixel) accuracy in
height can be attained with MOMS-02. The
drawing of further definitive conclusions
regarding the influence of the difference
in image mensuration quality between the
HR and LR imagery, and in the control
point distribution, number of orientation
images and order of polynomial
interpolation functions is unfortunately
precluded due to both the quality of the
ground point identification and the
apparent presence of the unmodelled
systematic errors alluded to by Kornus et
al (1995).
7. REFERENCES
Ackermann, F., Bodechtel, J., Lanzl, F.,
Meissner, D. Seige, P. and
Winkenbach, H., 1990. MOMS-02 - A
multispectral stereo imager for the
second German Spacelab Mission D2.
Int. Arch. Photogrammetry & Remote
Sensing, 28(1): 110-116.
Dorrer, E., Maier, W. and Uffenkamp, W.,
1995. Analytical kinematic sensor
orientation of MOMS-02 linear stereo
imagery, Proc. of Integrated Sensor
Orientation: Theory, Algorithms, and
Systems. Colomina/Navarro (eds.),
Wichmann Verlag, pp. 261-273.
—Ó——
TUR
