[he first
th-based
rface of
on two
is work
histori-
military
this has
nalogue
nic and
d into a
' 1980s.
ed from
ened an
tion of
lite over
art.
10rmous
multiple
4 to 8
)cessing
moving
over an
led into
onstruc-
sensor.
itioning
low-cost
0 stereo-
gle pass
mum of
s Wings)
an accu-
€ major
need to
nterfero-
xtended
|o good
1e direc-
g deter-
sensing
at from
iterest in
atellites.
ced for
needed
recisely
> greatly
ry from
'chnique
ocessing
tain an
een two
RS-1 and
r an area
her than
re globe
ıbject to
ollecting
rom the
satellite. The processes for repeat-pass interferometry
are fairly straight-forward. As a result there are many
centers of excellence that are capable of operating with
this technique. The difficulties are in those areas where
no useful information can be obtained. This may be in
areas of image lay-over or image shadows, and it is in
areas where the surface reflectivity is mirror-like (with
very smooth surfaces, where no echoes are being sent
back in the direction of the antenna) or where the
coherence of the echoes is marginal or fully destroyed,
which may be the case where excessive volume
Scattering occurs.
However, successful interferometry has been shown
even in cases that only recently were considered to be
highly risky. This includes mapping the surface of ice
sheets and, by differtial interferometry or by ex-
ploiting the geometric shape that may be known from
other sources, the motion of ice sheets can be
measured.
The accuracies of repeat-pass interferometry from
satellites can be defined by the elevation sensitivity
of the radar system and that is in the range of £2 m,
given an 12.5 m pixel. The accuracy itself is much
lower and is a result of the uncertainities in the orbit
and in the interferometric base vector. As a result one
typically has accuracies reported in the range of £15 m
Or SO.
The best of all options would be a single-pass satellite
interferometry system. This is being envisaged for a
reflight of the Space Shuttle’s SIR-C sensor. The
highest accuracy would be obtained if the uncertain-
ities about the base vector could be eliminated. That is
the case if two antennas are carried on one spacecraft.
The idea of a SIR-C reflight on board of the Space
Shuttle has been approved by NASA; the timing of the
interferometric reflight to cover the entire globe with-
in a few days by interferometric observation is still
uncertain.
7 USING RADAR IMAGE LAYOVER
7.1 Layover
An example of an very subtle and difficult element in
radar surface reconstruction is the problem that from
satellites the look-angle off-nadir is fairly small and
ranges from 20° to 45°. As a result areas with high
mountains where the elevation information is of
greatest value are laid-over. Layover is an effect that
occurs when the incidence angle is smaller than the
slope. In steep terrain where mountain sides may have
inclinations of 30° or more and rocks may be much
steeper than that, layover in images is a frequent
occurence.
The images in the layover are useless. Interferometry is
not applicable. Stereo-viewing is also marginal. As a
result vast areas of the mountainous surface become
non-observable from satellites, since they need to
look with small incidence angles.
7.2 Using the Layover to Define a Slope
To overcome this problem one can develop techniques
to automatically identify layover areas in radar
images. This is possible when multiple images exist,
425
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
e.g. from stereo-coverage at different incidence angles.
Layover areas are typically bright since they are facing
the antenna. A feature that is laid-over is identifiable
from stereo-coverage because it is wider in the steeper
looking image than in the shallower looking one. A
non-laid-over slope would manifest itself in the oppo-
site manner. Knowing that a feature is laid-over in one
image, namely the steeper one, leads now to two
potential solutions for the slope, depending on
whether the second image is laid-over as well. That
cannot be decided unambigiously. An opposite-side
view from an ascending orbit (if the stereo-views were
from descending orbits) can help to resolve the
ambiguity. Techniques to automatically employ lay-
over to refine a topographic model of the terrain are
currently being developed. A major application would
be on planet Venus. However, also in highly moun-
taineous terrain on Earth it might be applicable to
ERS-imagery.
8 FUSION
8.1 The Basic Idea
Fusion is a novel concept in machine vision that
suggests that multiple image and multiple methods
are being used in concert to obtain specific informa-
tion of interest. The implementation of this basic idea
is very applications-dependent. An industrial setting
may employ a vastly different solution to that which
may be useful for satellite remote sensing. The com-
bination of shape-from-shading, structured light,
stereoscopy, ranging, etc. may be considered a toolbox
in which the industrial robotics environment will
extract a combination most appropriate for specific
domains. The satellite remote sensing application may
use a combination of shape-from-shading, stereo-
scopy, interferometry, and exploitation of shadows
and layover and apply this to a suite of image
coverages of a given terrain in the optical and radar
sensor domain.
8.2 A Strategy for the Fusion of
Reconstruction Methods
Given the application-specificity of fusion any com-
mon strategy will have to remain on a very generalized
level. To take the example of topographic surface
reconstruction from satellite sensors we propose that
the methods of stereoscopic measurements, shape-
from-shading, and interferometry be combined with
optical surface classification.
The procedure would need to match the individual
images to one another so that an object area’s pro-
perties are reflected in a set of identifiable pixels in
the image. One therefore would want to exploit geo-
metric disparities in the stereo process to create first a
DEM, then a stack of co-registered terrain-corrected
images. The DEM used for the co-registration could be
obtained by interferometry. Therefore interferometry
and stereo might complement one another in areas
where one would work and the other not. Once the geo-
metric disparities are removed, shape-from-shading
can now be used on that image stack to extract further
detail of the surface geometry and radiometry. Those
properties may be obtainable by classification from
optical data and may then be verified and detailed by
the microwave observations.
