The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B4. Beijing 2008
resolution). Then processed CTX DTMS and ORIs based on
HRSC photogrammetic data can be employed as the source of
the geodetic control for «lm HiRISE stereo images.
Considering the significant resolution differences between CTX
and HiRISE (a factor of ~30), the registration is performed
hierachically between CTX products and resampled HiRISE
images of up to 3-4 metres resolution and then the intersection
points and ORIs of the resampled HiRISE are feed-forwarded
into the next finest resolution. The fact that CTX and HiRISE
data are acquired simultaneously on the same platform also
helps considerably. Such processing reduced the possibility of
the wrong co-registration being accidentally found and
gradually improves the positioning accuracy of the topographic
products by the triangulation process at each stage.
The other important approach in our work is the employment of
non-rigorous sensor models which was first introduced by Kim
and Muller (2006). These sensor modelling methods provide an
unified method to be applied with different sensors and can
readily be incorporated into our geodetic control strategy.
The ideal scenario of stereo processing is to employ HRSC -
CTX - HiRISE multi-resolution chains but the number of cases
that a target area includes all three stereo image sources is
relatively rare. A more direct registration between HRSC and
HiRISE or MOLA and HiRISE was also attempted. Then, it
was shown that this approach can be successful if the base data
has an acceptable positional quality and resolution even though
the photogrammetric accuracies of these products are not
comparable with the ones processed using the stereo HRSC-
CTX-HiRISE chain. The detailed photogrammetric accuracy
check of our stereo processing line is treated in a separate study
in depth (Kim & Muller, PSS, in preparation).
2.1.1 HRSC DTM extraction
HRSC stereo DTMs are a very important source of global
topographic slope data as well as geodetic information in this
study. Since the accuracy of the extracted local surface
roughness and other data’s photogrammetric quality are largely
dependent on HRSC DTM quality, a processing scheme to
produce relative noise free intersection point clouds has been
developed. In our stereo scheme, a two stage image matching
system was employed. A front end image matcher based on
Zitnick and Kanade (2000) algorithm generated high density
seed points for ALSC (Adaptive Least Squares Correlation)
refinement. The subsequent 3D intersection points from this
matching scheme and the HRSC sensor model are compared
with MOLA to remove significant outliers by slope analysis.
Then the median filtered values at 50m resolution, which is
around one-third of the size of MOLA footprint, are extracted.
Such a processing method provides a natural sloped height
surface so that the calculated local roughness values based on
this DTM are relatively free from the influence of any height
outliers and uncorrelated with slope. The resolution limit of a
re-constructed DTM by this processing chain is actually around
25m with 12.5m original image resolution if the stereo image
quality is relatively noise free. However, for the geodetic
control of other stereo imagery, the most reliable intersection
values in each matching position are chosen as the GCPs for
other image’s geodetic control. Then reliable control
information at 12.5m resolution can be achieved.
2.1.2 HiRISE DTM extraction
Currently, the most common stereo coverage of Martian surface
is HRSC imagery. However the quality of reconstructed 3D
Martian surfaces from HRSC stereo pair with 12.5 m spatial
resolution is still not appropriate for some geological
applications. For example, for direct surface roughness
extraction. The successful deployment of the NASA MRO with
the 25cm HiRISE instrument provides an opportunity to address
this issue for up to 1% of the Martian surface. One of the
difficulties in fully exploiting the potential of HiRISE for
photogrammetric products is that there are some technical issues
to use the tracking information for sensor modelling (Kirk et al.,
2007). Therefore we have developed a simple workaround
mapping method for HIRISE imagery using the strategy which
is described here. This appears to produce very reasonable
quality sets of mapping products including 0.5-4m resolution
stereo DTMs and 25cm ORIs over various HiRISE stereo
observation areas. If there is no CTX coverage, to enable
precise registration between 12.5m resolution HRSC and 25cm
HiRISE image, the scheme was constructed using a hierarchical
processing chain and by combining an efficient image matcher
which exploits both epi-polarity and ALSC (adaptive Least
Squares Correlation; Gruen, 1985) refinement methods at each
stage. Also the image matching noise and sensor characteristics
are effectively removed by a multi stage iterative surface
matching method. Therefore, this method can produce DTMs
with a resolution from 0.5-4m according to the processing stage.
2.1.3 CTX stereo DTM extraction
CTX stereo DTM has only been discussed in one short report on
the NASA Ames stereo pipeline (Broxton and Edward, 2008).
We have shown that our geodetic control method works well
with CTX stereo images and produces DTMs which have 12-
18m resolution. The stereo coverage of CTX is not yet as good
as the HRSC stereo coverage but compared with HIRISE, it
provides the higher resolution DTM availability at a reasonable
resolution. The other importance of CTX stereo topographic
data is its role to link between HRSC and HiRISE processing.
We applied almost the same processing method as for HiRISE
processing which is described here. It appears to produce good
quality DTMs and ORIs probably because of the more stable
sensor structure which employs one CCD line array rather than
the complicated 24 CCD-TDI combination in HiRISE. Then the
products of the CTX stereo chain are employed to produce
geodetic control for HiRISE imagery.
2.2 MOLA data processing
Garder (1982) analysed the influence of various effects on the
laser pulse width as follows
_. , 2 2, 4Tar(§)cos 2 S n
E(a, ) = (a, 2 + o h 2 ) + -T—^+
4 R
-tan 6,(tan 6, + tan (0 + S„) +
c 2 cos 2 (0 + S„) (0
tan 2 S, cos 2 S,
cos 2 (<p + S„)
*■)
where O t is the rms laser pulse width, O h is the rms width of
the receiver impulse response, Rm is the range from the
spacecraft to the illuminated spot, 5^, is the surface slope
parallel to nadir direction, S ± is the surface slope to the nadir
direction, 'O'j is the nominal divergence angle, c is the light
speed and tp is the off nadir angle.
In equation (1), the first term represents a system effect, the
second the influence of height undulation and the third term is
the effect of slope. Usually the off nadir angle is almost zero
and the nominal divergence angle is relatively small, so (1) can
be simplified to