"USH
play an
mmetry
jeome-
| will be
10tos and
oncerning
costs and
e scanner
rease the
CCD lines
ction and
jr example
id velocity
when all
platforms.
correction
rammetry.
age pair.
s imaging
se relation
geometric
grey level
metric on-
important
RNU) and
th respect
to on-line
adiometric
ference of
e points is
parts you
on images
V and the
24 1995
15
amplitude of angular motion and
the relation of integration time
and CCD-clock with the cut- off
frequencies of the disturbances.
We assume a terrain,
characterised only by its
maximum terrain height z_, and
an isotropic distribution of height
variations. Then it is possible to
discuss all effects of the real
imaging process only using the
projection of the image lines on
the plane zz z,,. The ideal
imaging is done with a constant
flight velocity (x-direction), with a
constant altitude H. Each
exposure of the CCD-line in a
vanishing integration time
produces an image of the
underlying terrain with central
projection. The extension of the
imaged terrain cut in flight
direction B depends on the
altitude H, the terrain heights z,
the focal length f and the width of
the sensor line dp.
B(z)=(H-2)*dp/f (1)
B is varying in swath direction
because z is varying. B is minimal
for z= z__. The ideal scan must
not have gaps. This is true if the
sensor moves between two
exposures over a distance B
Fig. 1 Nadir and forward looking images without geometric corrections
H=4000 m , Harz Mountains Bz(H-z,, )* dp/f (2)
Of course this sampling produces oversampling if z < z___ but this has to be considered in a resampling procedure.
This is the best scan under the made assumptions. The images are central projections onto the plane
z=z__. So, all disturbances can be discussed analysing their consequences in this projection plane.
A nadir looking sensor line is placed at the centre of the focal plane parallel to the y-axis. A stereo line is shifted by the
distance d in the focal plane. This produces the stereo angle o:
tan(o) = d/f (3)
The stereo images differ only by geometrical reasons even in an ideal scan. The parallaxes make differences and if the
ratio d/dp is not an integer value, we have a shift between the two ideal lattices of projected scan lines. The gradient of
terrain height changes the images too. The extreme case is a slope near by d/f .
Real scanners have a finite exposure time. This produce a motion blur. Modern scanners can change the integration
time. This has the advantage, that the right integration time can be found for example as an optimum value comparing
motion blur and s/n ratio and it can be tuned with respect to the illumination conditions (Jahn/Reulke, 1995).
The consequence of finite exposure times is, that camera position and velocities influence the imaging process.
At least 12 quantities are necessary to describe the geometry of imaging process: Three attitude angles, the vector of
angular velocity c, the vector of camera position R and the vector of linear velocity v. The six degrees of freedom lead
to other degrees of freedom of a vector r in the projection plane: two components of translation, one angle of rotation
and the new degree of freedom: The dilatation. Each pixel can be characterised by its viewing direction vector b :
b = (d, i*dp,f)" i=-(N-1)/2....0.....+(N-1)/2 N = number of pixels (4)
This vector is given in the camera fixed co-ordinate system. (z-axis down). The pixel position in the focal plane is given
by its number multiplied with the pixel spacing dp. In order to calculate the positions of each vector in the projection
plane the following vector equation can be used:
IAPRS, Vol.30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995