3. Istanbul 2004
for removal of
cking in ISAT.
)W POINTS
in the image,
es. The most
ent around the
1 abrupt change
dow is cast by
od for detecting
based on these
ts
d the first is the
shadow usually
| objects, points
gradients. The
ted as shown in
gradient of the
gonal directions
(1)
itensity gradient
onal directions.
calculated, it is
t is selected as a
> threshold.
buildings. There
x buildings that
be detected by
area around the
Model (DSM):
around the point
The area should
1adow. Once the
are matched by
area may not be
g is used. The
uted by forward
nages from first
: a local DSM is
^d to the created
a small area is à
ibrupt change in
, buildings, etc.
rst-order surface
squares method.
The standard deviation of elevation difference between points
and the fitted surface is then computed, which represents the
change of terrain slope in the area. When the standard deviation
is larger than the given threshold, it can be determined that
there is probably a tree or a building in the area and point is
removed as a shadow point.
Elevation
4—— A tree
o».
Lad
Figure 3. Slope Change in the Area with Trees
4. SOME EXAMPLES
The developed approach has been tested within Z/I Imaging's
ISAT and some test results are presented in this section. The
project presented in this paper consists of 4 strips and each strip
Qi iir
c Y
i
ied 0d
ed 1L ^
s CMS ra 5
à
m^
n
(c)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
has 7 images, taken with analog aerial camera. The images have
a scale of 1:6800 and were scanned at 14um. There are a lot of
trees and shadows in the project area. The project was run with
and without the function of removing shadow points. Figures
4(a) and 4(b) show the tie/pass points of an image in the project
generated by ISAT before and after removal of shadow points
respectively. It can be seen that plenty of tie/pass were created
cross the image before removal of shadow points, among which
some are good points, such as corners of sidewalks. But, many
shadow points and points on top of trees were generated at the
same time since the main features in this area are houses, large
buildings, trees and shadows cast by these objects. With the
function of removal of shadow points, most shadow points were
removed successfully. Figures 4(c) and 4(d) show the points in
a small area (black boxes) before and after removal of shadow
points. Totally there were 12 tie/pass points generated in this
area, seven of which are shadow points. After removal of
shadow points, all shadow points were removed and there were
eight points generated, as shown in Figure 4(d). Almost all of
these points are new points since the algorithm removed some
non-shadow points incorrectly and added some good points
automatically. Another example is shown in Figure 5.
(d)
Figure 4. Automatically generated Tie/Pass Points (a), (c) extracted tie/pass points before removal of shadow points, (b), (d)
extracted tie/pass points after removal of shadow points
93