lereo
JEM
] the
78m
e the
ined.
) an
| this
tereo
is an
hich
lares
tereo
(day
r can
ints,
n the
takes
hing
first
odes
991)
ellite
lieve
ihe
> the
The
)pler
1) to
(2)
(3)
(4)
32 is the velocity of the sensor for image2
$1 is the position of the sensor for imagel
32 is the position of the sensor for image2
) is the radar wavelength for imagel
).2 is the radar wavelength for image2
Pis the velocity of the target point on the ground
P is the position of the target point on the ground
In the above four equations, the sensor position and
velocity vectors are provided by the header data file in
each image. And 3D coordinates of P are unknowns
which are solved by the Least Squares iteration
technique. From a geometric view, (Curlander,1984)
noted that this approach is determined by three faces (1)
Earth's shape (2) Doppler equation and (3) range
equation. That means, at a particular time, the range
equation determined the surface of a sphere, while the
Doppler equations described the surface of a cone, the
intersection surface of a sphere and a cone yields a circle
which is intersected with the Earth model and to give the
exact position of a target point.
It should be noted here, that this intersection procedure
must be accomplished on the inertial reference system
with respect to geocenter. Because the two overlapping
images may be taken at different times, the inertial
coordinate system for each image may be also different.
It is necessary to convert one system to coincide with
the other. The conversion factor is related to the GMST
(Greenwich Mean Sideral Time) of the system. In order
to carry out the intersection we require matching results
and header data to provide geocentric coordinate for each
terrain point. The intersection procedures include many
steps of calculation which are illustrated in the
intersection flow chart figure 1.
4 EVALUATIONS OF STEREO MATCHING
In the CHEOPS algorithm, there are many parameters
that affect the matching accuracy, from the parameters of
generation the seed points to the image tiers used in the
matching. In this paper, the evaluation of the matching
is based on DEM accuracies. There are four different seed
points sets which are used for assessing the matching
results for several aspects, which will be discussed in
detail separately.
4.1 Data set
The test data in this paper include one ERS-1 precision
image (center incidence angle 23°) and one ERS-1 rolt-
tilt mode image (center incidence angle 35°). This
overlapping area covers Marseilles and Aix en Provence
in south France. The DEM in this area is also available
generated by IGN but part of the overlapping area is not
covered by the ground truth. For evaluation of DEM
accuracy, 512*512 portion of the overlapping area is
extracted.
4.2 DEM accuracies and single image tier accuracy
For the CHEOPS PDL file used in this research the
matching results from the preceding tiers are multiplied
by the factor 2 and used as initial values in the next tier.
Because of this the total DEM accuracy can be estimated
by a single image tier. Table 1 illustrates this fact.
381
read header data read header data
from imagery L from imagery R
preliminary calculation
+
coordinate translation (screen->image)
t
prediction of orbit position & velocity
position & velocity vectors
coordinate transformation
(geocentric -»inertial)
intersection
unknowns solution
coordinate transformation
(inertial -> geocentric)
Figure 1: Intersection flow-chart
In table 1, six image tiers wee used, the sixth tier is is
the original image, the fifth tier is reduced by 2 And so
on. From this table it is obvious that the total DEM
accuracy is strongly influenced by the fifth tier results,
for there are about 80% of the matching points on this
tier. Also, the table showed that for the ERS-1 SAR
imagery, it is better to undertake the matching on the
lower resolution of the 5th tier image rather than the
original one, this is shown in row 10 that DEM accuracy
of tier 6 is lower than any other tiers of image. This also
illustrates the benefit of using CHEOPS.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
