Geocoding Based on Recovery of Image Acquisition
Parameters
In this approach, ground control points described by
three-dimensional object coordinates are transformed into
the image space. Differences detected by location checks
using the homologue image points are taken as parameters for
a preliminary adjustment of imaging characteristics, in
particular of sweep delay, time off-set and scale factors,
Then, in analogy to the photogrammetric approach,
measurements of image coordinates of control points and the
SAR projection relations can be combined and optimized in gj
complex bundle adjustment procedure leading to a restitution
of ephemeris data. Thus, the underlying radargrammetric
model for the final image-to-object space transformation is
adequately adapted to provide a valid description of the
original image geometry.
It is evident that, within limits, both geocoding
methods are able to tolerate only approximately given
ephemeris data. In different ways, they both recover
consistency of the radargrammetrio model with the geometric
distortions appearing in the input SAR image. However, a
comparison of advantages and draw-backs of these two
algorithms would go beyond the scope of this paper.
Image-to-map rectification offers the possibility to
integrate SAR image data into geographic information systems
(cf. Howard 1985) and to efficiently convert mosaics of
geocoded SAR images into radar image maps. Change detection
for topographic and thematic map revision and the correction
or concentration of information in geodata bases are other
applications. Moreover, in many cases it is a prerequisite
to the registration of multi-sensor remote sensing data.
3. STEREOSCOPY
In principle, radargrammetric fusion of two SAR images
covering the same area but referable to different image data
acquisition parameters enables the reconstruction of
three-dimensional object coordinates. Two such images from
a SAR image stereo-pair. However, the sensor position
distance must not be too long (cf. Raggam et al. 1985).
Terrain information of both images is transformed into
the object space. In doing so, attention must be paid that
each of them is transferred using the respective sensor
model. The intersection points of any two range projection
circles passing through identical points in the two images
are determined, usually in a least Squares adjustment
calculation. The coordinates of these intersection points
represent location and elevation of the imaged object
points.
326
ar