The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1296
A final report on the DLR/IMF results achieved for CSAP
together with a set of DSM and orthoimages has been delivered
to ISRO/SAC. A copy of the report can be provided to
interested readers.
1.2 DLR test sites for CSAP
DLR/IMF takes part in CARTOSAT-1 Scientific Assessment
Program (C-SAP) as a principal investigator for German
(Southeast Bavaria) and Spanish (Catalonia, test site 10) test
sites, and as a Co-I in the evaluation of CARTOSAT-1 data for
test site 5 (Mausanne-les-Alpilles, France). In all cases rational
polynomial functions (RPC) are provided by the distributing
Indian agency ISRO/SAC as a universal sensor model for each
scene. Table 1 shows the imaging dates and the centre roll angle
which is varying substantially (e.g. for Mausanne scenes for
producing the overlap).
Abbreviation for
the paper for aft
and fore scenes
Imaging date
Centre-roll (deg)
(for aft image)
Mausanne-les-Alpilles (France), test site
5
MAI /MF1
31Jan2006
-13.6
MA2 / MF2
06Feb2006
4.0
Catalonia (Spain), test site 10
Cat-A / Cat-F
01Feb2006
-0.1
Bavaria (Germany)
Bav-A / Bav-F
30Apr2007
9.5
Table 1 : The 4 stereo pairs used for C-SAP at DLR
Some numbers in the tables given in this paper may be slightly
different from those in (Lehner et al. 2006/2007) because the
original RPC delivered with the images had zero denominator
problems and were later replaced with new RPC by ISRO/SAC.
2. DLR STEREO PROCESSING
2.1 Image matching
Hierarchical intensity based matching as implemented into the
XDibias image processing system of DLR/IMF consists of two
major steps (Lehner and Gill 1992; Komus et al. 2000). In a
first step the matching process uses a resolution pyramid to
cope even with large stereo image distortions stemming from
carrier movement and terrain. Large local parallaxes can be
handled without knowledge of exterior orientation (which is
often not available with sufficient accuracy for space-borne
imagery). The selection of pattern windows is based on the
Foerstner interest operator which is applied to one of the stereo
partners (chosen according to the best radiometric properties -
in case of CARTOSAT-1 this is the aft image). For selection of
search areas in the other stereo partner(s) local affine
transformations are estimated based on already available tie
points in the neighborhood (normally from a coarser level of the
image pyramid; on the coarsest level (factor 64 reduction in
case of CARTOSAT-1 stereo pairs) the parallaxes and shifts are
already so small that the process can be started automatically
just using adapted (larger) window sizes for patterns and search
areas). Tie points with an accuracy of one pixel are located via
the maximum of the normalized correlation coefficients
computed by sliding the pattern area all over the search area.
These approximate tie point coordinates are refined to sub-pixel
accuracy by local least squares matching (LSM). The number of
points found and their final (sub-pixel) accuracy achieved
depend mainly on image similarity and decrease with increasing
stereo angles or time gaps between imaging. The software was
originally devised for along-track 3-line stereo imaging (stereo
scanners MEOSS and MOMS operated by DLR). Normally, the
procedure can be executed fully automatically if the shift
between the stereo partners is small compared to the image size
as is true for CARTOSAT-1 stereo pairs. The procedure results
in a rather sparse set of tie points well suited for introducing
them into bundle adjustment and as an excellent source of seed
points for further densification via region growing (second step).
The second step uses the region growing concept first published
by Otto and Chau in the implementation of TU Munich (Heipke
et al. 1996). It combines LSM with a strategy for local
propagation of initial conditions of LSM.
Various methods for blunder reduction are used for both steps
of the matching:
• Threshold for correlation coefficient
• Bi-directional matching and threshold on resulting
shifts of the coordinates
• Quasi-epipolar reprojection of tie point coordinates
In areas of low contrast the propagation of affine transformation
parameters for LSM in the region growing process leads to high
rates of blunders. In order to avoid intrusion into homogeneous
image areas (e.g. roof planes and agricultural fields without
structure) the extracted image chips are subject to (low)
thresholds on variance and roundness of the Foerstner interest
operator. This and the many occlusions found in densely built-
up areas imaged with a large stereo angle create lots of
insurmountable barriers for region growing. Thus, for high
resolution stereo imagery the massive number of seed points
provided by the matching in step one (image pyramid) turns out
to be essential for the success of the region growing.
The numbers of tie points found and their sub-pixel accuracy is
highly dependent on the stereo angle. A large stereo angle
(large base to height ratio b/h) leads to poorer numbers of tie
points and to lower accuracy in LSM via increasing
dissimilarity of the (correctly) extracted image chips.
For currently available high resolution stereo imagery the stereo
angle is too large, at least for built-up areas. The importance of
a large base-to-height ratio is exaggerated at the cost of the
matching accuracy and density (see Krauss et al. 2006). The
accuracy in forward intersection is inversely proportional to the
base-to-height ratio but also direct proportional to the matching
accuracy (parallax measurement). The latter and the matching
density are improved by reducing the stereo angle. In table 2
past and current stereo missions are shown together with their
stereo angles. In contrast to the drastic reduction in ground
sampling distance (GSD) the stereo angles are growing instead
of decreasing. Thus, image matching performance is much
hindered because of the appearance of more and more complex
natural and man-made objects in the images. For IKONOS-2
the possibility for smaller stereo angles exits. As indicated in
the table 2 DLR managed to get examples with 10 and 6 degree
stereo angles and the advantage for image matching could be
shown (Krauss et al. 2006). An increase in the accuracy of the
parallax measurement via finer resolution allows for a decrease
of the stereo angle to improve matching performance.