a track at an
in the row i
point X; on
ns
(3-3)
on an ideal
es
erical order
rcraft in the
CD-line will
The true height at the measured point can be calculated
with (4-5) and the known attitude parameters (4-6).
The height error affecting only the row difference of the
corrected image is
Ahb 8X.
— = 1-— (4-7)
h 6X4
5. STEREO RECONSTRUCTION
The following procedure was used for the reconstruction of
a digital elevation model (DEM) . The first step is finding
conjugated points in at least two different images using a
matching algorithm. With the knowledge of their coordi-
nates in the focal plane and of the attitude of the sensor,
so-called pixel rays can be defined.
Xin = Xog tt Xqq (5-1)
Xon current camera location,
Xdn camera's direction vector,
t unknown parameter,
where n is the number of the current line.
Under ideal circumstances the three coordinates of an
object point in the terrain are given by the intersection
point of these rays. Because of the discretization errors
caused by the finite resolution of the camera, there is no
intersection point. So an error criterion must be defined to
determine the vector with the smallest distance between
the rays. This vector gives the 3D-coordinates of the
reconstructed point.
3 0 0 1 1
Ki =X, tly X75 =X, 70 X75, (5-2)
Xi estimated intersection point
$ deviation between estimated and real inter-
section point.
Equation (5-2) demonstrates the idea for two rays. The
estimated intersection point between the rays and the ter-
rain is defined as the vector, where the error is minimal.
Figure 6 shows this approach for a CCD-line scanner.
flight direction
ngl
a | stereo angle
terrain model
Figure 6 3D-coordinates of the reconstructed point
After the matching procedure pixels of different CCD-lines
689
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
are related to the same object point. Pixel rays were
defined and the three coordinates of the point in the object
space where obtained. If a sufficient number of terrain
points could be calculated, a DEM can be retrieved by a
two-dimensional interpolation algorithm. This procedure
was executed with simulated image data and yielded
excellent results[4].
Figure 7 shows the different ways of generating digital ele-
vation models using the received image data.
original image data
correction |
EN
matching
%
real flight data
r4
digital terrain model
ideal flight data
4
Figure 7 Ways of DEM generation
The first possibility is to correct the disturbed image data
as shown in a previous section. The result is an image file
without any relevant effects caused by the flight motions,
like blurs and pixel shifts. The ideal attitude data can be
used for defining pixel rays. After that a DEM can be built.
A disadvantage of this procedure is that the correction
algorithm can cause height deviations in the DEM as
derived in equ. (4-7).
The second way is matching the original, uncorrected
images to find conjugated points, and using the original
attitude data to define the pixel rays to produce a terrain
model. Because there is no correlation between the atti-
tude data of two image strips (the baselength is too large)
it gets harder for the matching algorithm to work correctly.
The number of retrieved conjugated points decreases with
increasing flight motions and depends on the terrain under
observation, of course.
So the best way should be to match the corrected images
first to obtain the maximum number of conjugated points
and to define the pixel rays with the help of the original atti-
tude data. It is just necessary to remember the original
pixel position in the focal plane before the correction.
Figure 8 shows the result of the described procedure. Two
image strips of a flight over Ronneburg were evaluated
and a DEM was generated. In Figure 9 the corrected nadir
image strip was laid over the DEM.
