The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
For the new approach the elimination of the pitch influence is
difficult, because the effect depends on the height of the objects.
That means, the higher objects stand out from the reference
plane, the more precise navigation for the aerial survey must be
carried out. If the pitch angles of the scan lines are close to zero
then the displacements perpendicular to the line will be smaller
and the true ground coordinates become more precise.
The roll angle shows relatively quick and sudden changes in the
raw image data and in the diagram. However rolling affects the
new approach rather less. Thus the image points are also shifted
here, but along the scan line. The influence has its maximum on
the margin of the image strip, and is also depending on the
height of the objects. However, this does not cause problems for
orthoimaging, because the final ground coordinates are taken
only perpendicular to the scan line (in flight direction).
The yaw angle refers to the north direction, thus it is different in
the absolute values for both flight strips (see Figure 4 below). In
general the yaw angle effects an azimuthal rotation of the
original scan line and leads to deviations in x- and y-direction.
But the x-direction (perpendicular to the scan line) is more
important, because in this direction the objects are shown in
parallel projection. The differential changes of the yaw angle
from scan line to scan line cause that the flight direction does
not agree any more with the x-direction and that the x-
directions of the original scan lines change in the corrected
nadir image constantly, e.g. by a drift.
However, a constant x-direction is essential for the new ortho
imaging approach, because the correct ground coordinates are
taken in this direction. If a constant x-direction is defined for
one image strip, e.g. by an average yaw angle, angle differences
appear in every line between constant and real x-direction. The
angle differences lead to shiftings perpendicular to the scan line
with a maximum at the margin of line. The shifting becomes
too large and one gets incorrect ground coordinates, by
processing the complete image strip with one average yaw
angle.
To solve this problem a subdivision of the overlapping image
range into smaller segments with other x-directions seems
feasible to hold the influence by the deviation of the individual
x-direction low. As an example, the images in Figure 3 have an
extent of 1000 pixels and a maximum angle difference of 0.2
degrees. Thus the difference yields incorrect ground coordinates
by 3 pixels at the end of the line.
The differences between the yaw angles of both flight strips
contain also possible effects of deviations from the ortho
gonality between the flight lines. Also in this case the dif
ferences are not constant from line to line. The deviations have
its maximum on the strip margin and the effects are along the
flight line. For the image segment in Figure 3 a constant dif
ference was defined, thus the general part of the deviations can
be attached to the incorrect ground coordinates, by a rotation.
For the data investigated an average deviation from the ortho
gonality was determined as approximately 1.7 degree. This
leads to a correction of approximately 17 pixels across to the
line.
4.3 Occluded Areas
A general problem of true othoimages are occluded areas
behind buildings, bridges, etc. Rather sophisticated techniques
must be applied to fill the resulting gaps in the images. The
situation will be improved due to the mixed projection of
pushbroom scanners, where relief displacements occur only in
one direction. Theoretical studies show that through this effect
occluded areas are significantly smaller in pushbroom datasets
than in images acquired in central projection. That means, the
necessary computations for the correction of occlusions and
filling gaps in the true orthoimage will be reduced.
5. DEFINITION OF CORRESPONDING POINTS AND
TRUE ORTHOIMAGE GENERATION
For general investigations about the realization of image
matching with real image data the following simplifications
were assumed:
• the data acquisition of the data occurred in nadir direction,
• the azimuthal orientation (yaw angle) of every scanned
line corresponds to the middle flight direction.
Furthermore following preparations were carried out:
• geometrical and radiometrical corrected image data were
used
• the averaged flight directions were fitted to the coordinate
axis’s so that the flight tracks were oriented in parallel
with the respective axes of coordinates of the image coor
dinates system.
These preparations and simplifications enable a direct work in
the scan lines and the determination of the correct coordinates
from the image contents.
For an exhaustive allocation of both images a correspondence
analysis must be carried out for every pixel of an image. For
detection of corresponding points in the two data sets the
proven process of the normalized cross correlation was applied.
For every pixel it delivers a statistical value, which shows how
well the grey scale value variation of the searched pixel and its
environment agree with the respective pixel of the other flight
strip. The value range reaches from -1 to +1, and +1 is 100% of
correspondence, 0 is no correspondence and -1 indicates an
inverted correspondence. For the investigation only correlation
values were used by at least 0.6 for the allocation of an image
point.
For the sizes of the template matrices, which contain the grey
scale values of the examined pixel and its neighbors, values
between 3x3 and 9x9 were selected. The correspondence
analysis was carried out in both directions. Pixels from the first
image was searched in the second one and vice versa. The grey
scale values of the found points have been averaged. Because of
the similarity of a many pixels often ambiguities led to wrong
results. In order to minimize this effect the range to be
examined (search matrix) must be limited.
In the direction of central projection, objects which will stand
out from the reference plane became displaced. In Figure 5 the
displacement of some points is sketched schematically for both
flight directions. The displacement of a point increases with the
distance from the flight line.
376