Dataset 2 - stillvideo images
For a pilot study on the applicability of high resolution
digital stillvideo cameras in aerial photogrammetry a
block of 5 x 10 images was taken with a Kodak DCS200
(8 bit monochrome 1524 x 1012 pixel sensor, format
14mm x 9mm, pixel
spacing 9um x 9um,
internal harddisk for 50
uncompressed images,
18mm lens) from a heli-
copter over a village in
Grisons. The average
image scale was
1:18'000, the average
overlap was 70% in both
directions, with large
deviations due to
handling difficulties
with the hand-held
camera in the unsteady
helicopter. The orientation data of the images was taken
from a self-calibrating bundle block adjustment after semi-
automatic measurement of the image coordinates of 36
signalized points. Checkpoint accuracies of 2cm for the
planimetry coordinates and 4cm for the height coordinate
were achieved in this aerotriangulation (Kersten, 1996).
Figure 5: DCS200 aerial overview image of a village
An average of 7000 points per image was extracted by the
Foerstner-operator. Due to the geometric quality of the
data a maximum tolerance of 10 micron (~ 1 pixel) to the
epipolar line was found to be appropriate for the accep-
tance of possible matches. Applying the epipolar line
intersection routine with the requirement that each point
has to be matched in at least 5 images, a total of 6860
object points could be reconstructed. The maximum
number of images a single points was successfully
matched in was 21. The average standard deviation of the
reconstructed points was 0.027/0.027/0.109m in X/Y/Z.
An contour line plot generated from the 6860 object
488
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
points is shown in Figure 6. Again, no post processing
was applied to the raw data provided by the routine. The
plot looks quite good in the surroundings of the village,
but rather chaotic within the village.
hele a S
Si
Figure 6: Urmein test area, 1m contour lines (orientation and
scale do not correspond to Figure 5) .
* *
Some problematic regions inside the village are shown in
parts of a digital orthophoto mosaic with contour line
overlay in Figure 7: As no information on buildings, trees,
etc. was a priori given or explicitly extracted by the tech-
nique, a large number of successfully matched points are
not situated on the terrain surface, but on objects above
the surface. Due to the characteristics of the interest oper-
ator and the loss of points caused by the probability of the
interest operator detecting identical points in all involved
images, however, buildings are often modeled only rather
incompletely. In some cases only one or two points of a
roof were matched, and terrain points at the foot of build-
ings are often missing.
I ^
x NZ
blunders
Figure 7: Insuffently modeled ehurch,steeple:
parts (orthophoto
with matched points
and 2.5m contour line
overlay)
Sa
# missing foot points
A tho
with
showe
which
the re
at lea:
remai
in at
points
Tabl
min
obje
In coi
retica
indice
stner-
and a
identi
In coi
techn
of coi
a con
multi
epipo
showi
follov
As fa
conce
by th
most
the p
multi
corre
than
apprc
ered .
mour
One
point
sitate