contingent upon the provision of clear,
unambiguous targets in the imagery. Only
one category of target was found which
adequately fulfilled these requirements,
this being the water surface area of
dams. Even these targets were somewhat
deficient in that most dams had changing
water levels and diameters of 30m or
less, which translates to only. .a- 2-3
pixel width in the off-nadir imagery.
Moreover, there was an insufficient
number of such targets.
A second category of targets, which were
Clearly visible in the imagery, were dam
embankments. Whereas, the centroid of an
embankment image was usually clear, great
difficulty was encountered in finding the
corresponding position. on the ground.
Errors of a pixel or more could be
expected in this target point
correspondence operation and it was not
always feasible to assess which were
'poor' targets. A last general category
of targets comprised road (few), track
and fence intersections, or more
correctly in the latter case,
intersections of graded tracks along
fencelines. In most cases these presented
reasonable image targets, but there was
the complication that the imagery was
acquired in 1993, with the ground survey
being carried out in 1994 and 1995. In
each dry season a program of grading
occurs in which roads, tracks and
fencelines are re-graded following wet-
season damage, not necessarily in exactly
the same location as in the previous
year.
Notwithstanding identification problems,
about 80 well distributed image-
identifiable ground control points were
established. Two GPS campaigns were
mounted to provide the necessary ground
truth data. The GPS survey technique
employed two base stations and roving
receivers, with an occupation time of 30
minutes at each point. Processing of the
data from the 122 baselines observed
indicated that a positional accuracy
(relative) of 10cm had been achieved.
To help alleviate some of the problems
with point identification a number of the
stations were re-occupied in the 1995
field campaign. Re-observation of these
points confirmed the quality estimates
for the GPS survey. A last phase of the
second campaign was the survey of a 16km
3-D profile along an image-identifiable
fenceline via kinematic GPS. This
heighting profile was established to
facilitate an evaluation of the precision
of MOMS-02 DTM extraction. Further
details of the GPS survey phase are
provided in Fraser et al (1996).
The last component in the establishment
of the Australian testfield comprised the
image mensuration stage. Multiple image
coordinate measurements were observed for
208
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
each of the GPS-surveyed ground points.
In addition to the set of observations
made in monoscopic mode on an Intergraph
ImageStation digital photogrammetric
workstation at The University of
Melbourne, independent measurements with
varying levels of image enhancement were
also made by MOMS-02 research groups in
Germany and Switzerland.
Following qualitative analysis, backed up
by the results of a process of 2-D image-
to-ground transformation, a subset of 56
3-ray points and an additional six 2-ray
points were deemed likely to display
measurement accuracies at the 1-pixel
(10um) level or better. The distribution
of these.points is shown. in. Figure 1.
Some 40 of the image points were
estimated to display a standard error of
better. than 0.5. pixel in..all . three
imaging channels, based primarily on
image quality. In this paper we consider
two such data sets, one from Melbourne
(three channels) and one from Dr E.
Baltsavias of ETH Zurich (forward and aft
channels only). In the context of the
target identification problems referred
to. ut is noteworthy that the RMS
discrepancy between these two
independently observed data sets was 0.7
pixels or 7um, which is a little higher
than desired.
3. MATHEMATICAL MODEL
The functional model adopted for the
exterior orientation/triangulation is a
form of the photogrammetric bundle
adjustment adapted to accommodate the
geometric conditions of three-line
imagery (Ebner et al, 1992):
f R(X =X) + Ry (Y= 1p) + Ry (ZZ) - MAX + My AY + My AZ]
X= Xp —
R(X = Xy) t RA(Y — Yo) t R(Z — Z9) [M Ax T MAY - M,AZ|
(1)
R3QC- X) Ra(' - X) € Ra - Z) - [MAX - MasAY T MsAz]
va” f i
R(X =X) + Q7 X) (Z7 Z) - [MsAX T MAY T Maz]
Equation 1 expresses the image coordinate
observations x,y as a function of the
following parameters: the elements of
interior orientation: x, yj and f; the
coordinates X, Y, and Z of the object
point; the exterior orientation elements
X77 20,905, Q5; and k, of the HR,- nadir-
looking lens; and the relative positional
and orientation elements AX, AY, AZ, Ao,
Ap, and Ax of the off-nadir sensor line
with respect to the projection centre of
the HR lens. The rotation matrix R is
obtained as the product .of rotation
matrices M(Ao, Ao, Ax) and D(w,, Pos Ko) -
In the standard along-track stereo
imaging mode, without cross strips, self-
calibration is not possible and thus a
number of the parameters forming the
extend
determ.
include
and th
Az<For
channe.
02/D2 «
21-45
six to
In lord
collin
a re-|
orient:
polynor
Quadra!
stereo
1989)
al, 419
of MO
polynor
propos:
ali, |
which
invest:
elemen
orient:
line i
then m
of sen:
intervi
QOIS. JF
given 4
P(1) =
where
