The International Archives of the Photogrammetrw Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1033
tion (5) - involving bilinear interpolation for surfels within a
particular DTM facet. See Gehrke & Haase (2006) for formal
derivation and further details.
R = R° +
<3R° X-X 0
SX z°-z 0
H-
SR° Y - Y 0
SY Z°-Z 0
dZ
(5)
The entire Facets Stereo Vision algorithm has to be carried out
iteratively, starting with the resampling using enhanced DTM
heights, until no significant height changes remain. It has been
proven effective to start up with a comparatively coarse model
(MOLA DTM resolution) and increase it step by step (Fig. 2).
3.2 Regularization Aspects
The basic algorithm becomes weak in smooth regions that com
prise only little texture or small contrast differences - areas that
frequently occur on the dust-covered Martian surface. Thus, the
adjustment has to be stabilized by additional assumptions. This
may be realized by introducing terrain-dependent restrictions to
surface curvature (Wrobel et al., 1992; Gehrke 2007).
While such regularization significantly reduces noise and over
comes DTM outliers, in particular when locally weighted by
surface properties, such conditions are only fictitious. This lack
could be overcome by fully using all radiometric image infor
mation, i.e., the observed radiance factors of each pixel, and to
combine matching in object space with photoclinometry.
Initial Heights
(MO LA)
HRSC
image Data
Г
Orientation
\
L.
Data
✓
о
£
О
«я
Ф
ОС
о __
Ф ге
«л <~
та Ф
ф а
Definition of
Facet Sizes
Differential
Rectification
i
r Pseudo >
^ Orthoimages J
- r ^—'
Atmospheric
Modeling
Atmospheric
Optical Depth
Radiometric
Adaptation
i
Photo-
* Matching in
clinometry
Object Space
Integrated Modeling Approach
*
t
>
DTM
J
Reflectance
Parameters
4
Orthoimages
Hapke
Modeling
~~r~
Material
Properties
Fig. 2: Integrated approach for geometric and radiometric mo
deling of the Martian surface.
4. THE INTEGRATED APPROACH
4.1 Combination of Facets Stereo Vision and Photoclino
metry
Radiometric image formation is described by a surface reflec
tance model combined with an atmospheric model (see 4.3 for
the latter). Assuming, the reflectance model and its parameters
are known, local surface normal vectors can be derived - see
equation (1). Since this equation is in the form of (5), it readily
allows for the integration of photoclinometry into Facets Stereo
Vision (5), once local surfel inclinations in (1) are expressed in
terms of dZ, again, with bilinear interpolation within the DTM.
Such an approach nicely combines the advantages of both mat
ching, which provides reliable absolute heights for well textured
image parts (global accuracy), and photoclinometry, which
leads to relative heights (local accuracy) and is suited to bridge
matching gaps. This goal is achieved by weighting the two al
gorithms in the combined least squares adjustment depending
on local image texture, practically carried out using radiance
factor gradients in X and Y that are already part of (5).
4.2 Radiometric Surface Modeling
In most cases, the assumption of known reflectance parameters
does not hold true; they have to be determined. Here lies the
major advantage of the combined approach: If matching leads
to (some) reliable heights that constrain inclination, reflectance
parameters may be derived within the photoclinometry part of
the adjustment. This ability requires certain relief energy as
well as good image texture.
However, the Hapke parameters, described in chapter 2, tend to
be highly correlated and the derivation along with DTM and or
thoimage will very likely fail. Therefore, the empirical two-
parameter Lunar-Lambert-Model - a suitable approximation for
Hapke, proven throughout various theoretical and practical in
vestigations (McEwen, 1991; Kirk et al., 2004) - is chosen as
the radiometric surface description in this context:
R = A
2L-
COS0:
COS 0: +COS0,
+ (l-L)cos0¡
(6)
The normal albedo A in (6) is linked to the particle single scat
tering albedo w, the limb darkening parameter L to the surface
roughness in the Hapke Model (1). Thus, based on the empirical
reflectance resulting from the adjustment, physically mea
ningful Hapke parameters can be retrieved afterwards (Fig. 2).
4.3 Atmospheric Modeling
For an appropriate surface modeling, attenuation of the atmo
sphere (optical depth) and the influence of ambient light are to
be regarded - cp. equation (5). While the latter can be integra
ted into the adjustment as an additional parameter, the optical
depth is derived in advance since the related multiplicative term
in (5) would highly correlate with the albedo factor in (6).
Optical depth can be derived by the Stereo Method (Hoekzema
et al., 2006), which uses HRSC stereo bands: different angles
result in different atmospheric path lengths and, according to
equation (3), different influences of optical depth, which causes
different image contrasts. Then - assuming identical ambient
light and allowing only linear differences in surface reflectance
between stereo angles - the optical depth could be derived from
contrast differences, ideally from nadir and the outermost stereo
band(s) as those feature the biggest path length difference.