983
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
Figure 2: Eauidistant man of current surface coverage of SRC images on Phobos.
4. CONTROL POINT NETWORK
4.1 Duxbury and Callahan control point network
The Viking orbiters were the first to obtain complete ground
coverage of Phobos with excellent phase angle coverage. Pixel
resolutions of the images are in the order of 200 m or better.
The first global control point network consisted of 98 points
(Duxbury and Callahan, 1989b), mostly craters. It was later
extended to over 315 control points. The measurements
included relatively large craters covering several pixels.
Uncertainties of the 3D-coordinates of the points ranged from
±74 m to ±900 m (Duxbury, 1991).
4.2 HRSC and SRC image data
The resolution of HRSC images from a range of 2000 km is
approx. 80 m. Hence, though the built-in stereo capability seem
to make the HRSC attractive for control point network analysis,
only a limited number of HRSC images from more or less 10
flyby maneuvers would be suitable for an in depth control point
network analysis.
For SRC on the other hand, with its large focal length of 988.5
mm, the ground pixel size is smaller by a factor of approx. 4.3
over that of HRSC. Pixel resolutions of SRC images for flybys
range from 100 m/pxl to 5m/pxl. The back draw in case of SRC
image data is, that SRC is a single frame camera, pointed to a
fixed point in the stellar sky during one flyby. Thus, the same
surface area needs to be observed during different flybys with
different viewing angles to be useful for photogrammetric
techniques. The establishment of an independent control point
network for Phobos requires a global image coverage. HRSC
and SRC images cover approx. 74% of Phobos’ surface in
stereo (cf. Figure 2). Viking orbiter images were incorporated
into the image database to fill the gap on the anti-Mars side of
Phobos. Fortunately very high resolution images, of the missing
area, are available in the Viking data set.
4.3 Object point determination
We measured line/sample coordinates of control points in 53
SRC images and 16 Viking orbiter images. Contrary to
Duxbury and Callahan (1989b) our control points are defined as
the centers of a crater on the crater floor. Since image
resolutions allow us to observe very small surface features -
even small features within large craters - we assume that the
surface features are shallow and represent a mean surface to
sufficient approximation. A total of 660 points were observed
3845 times with a minimum of 2 observations and a maximum
of 14 observations, but on average 6 observations per point in
both image data sets.
Initially the two sets of observations - SRC images and Viking
orbiter images - were processed separately to control the
predicted orientation information. Nominal navigation data for
the SRC images could directly be used to determine object
point coordinates of the control points in a least-squares bundle
block adjustment. For a few orbits, normalized residual values
of the camera orientation data indicated that uncertainties of the
position and pointing observations were larger than expected.
An adopted weighting scheme for observed camera orientations
was applied to reduce the relevance of uncertain observations
on the adjustment. Estimating that image coordinates were
observed with a 1 pixel uncertainty, mean object point
accuracies Ox, Gy, Oz of 39.6 m, 34.6 m, and 36.1 m,
respectively, were computed, for control points measured in
SRC images.
SRC
Viking orbiter
No. of points obs.
2953
871
Umax [m]
92.1
164.2
Omin [m]
32.7
91.7
Omean [m]
36.8
100.9
Combined Adjustment of SRC and Viking
No. of points obs.
3841
Omax [m]
62.5
Omin fm]
4.4
Omean [m]
15.9
Table 3: Object point accuracies for the different bundle block
adjustment models.