s 14
slant
se S
Osite
two
ners.
1 the
ased
8 99
0' e"
/
/
/ dh
/
/
/
/
/ dy
(P) ii 7
de se : /
YY
/
/
Figure 1: Definition of entities for propagation of range error into errors of height and cross-track
coordinates.
Equ. (7) is now easily seen to be identical to equ. (2):
Suis = Orange * ((sin® ©’ + sin? 6")12/(sin ©’ . sin 0")).
(sin ©’ . sin ©")/(sin ©".cos ©’ + sin ©’ cos 9") (7)
Table 2 lists the coefficients of equs. (2) and (4) for the various stereomodels that one can form from
the SIR-B data. All stereo-image pairs are from same-side parallel configurations except for the
Illinois-case which is opposite side. It is evident that the predicted errors are smaller with larger
stereo-intersection angles. Generally the height errors would not be larger than twice the errors of
slant range.
The relationship between range resolution and error of slant range is unclear. In the absense of
accepted statements on this relationship it is assumed that the standard error of slant range is 1/2 of
the range resolution value. This would amount to a O range of 7 m.
2.2 Accuracy of Model Set-Up and Point Positioning
The work with actual SIR-B images relies on the analytical plotter Kern DSR-11, equipped with the
radargrammetric software system SMART that has been discussed by Raggam and Leberl (1984).
Image pairs are inserted into the instrument, an image coordinate system is defined and model set-up
measurements as well as computations are performed to obtain a parallax-free stereo-model. The
computation consists of a so-called "bundle-solution" that relates image coordinates to ground
coordinates with the help of ground control points. This is followed by actual data collection of
individual points, contour lines and of planimetric detail.
Table 3 summarizes the residual coordinate errors after stereo-model set-up. It is immediately evident
that these accuracies are far less than expected from propagation of slant range errors. It was argued
in a first discussion of these values in the earlier paper (Leberl et al, 1986b) that this could be caused
by poor stereo viewability due to migrating edges such as illustrated in Figures 2 and 3 that show a
detail of the Gordón la Graza data. Edge migration in this case is partly caused by the stereo-
geometry expressed in parallax differences, and partly by the difference in illumination angles. In
featureless rolling terrain such illumination-induced edge migration could be significant. Figure 4
illustrates the concept with the help of a sinusoidal surface relief. At a given position x along the
profile, the slope A is:
tan À — a.b. cos (b.x) (8)
Where a is the amplitude, b the period of the sinusoidal relief.
335