33. Istanbul 2004
12 to 15 meters,
> laser strips. The
1gs or part of the
buildings were
lative orientation
. pulse was used
seen in Figure 2.
le roofs and flat
mall one-storied
ntation of many
ither parallel or
h caused some
ispected.
et al, 2003) was
tween two laser
d was originally
tion between an
of Figure 3. The
control points,
xample.
sed on visual
the image. The
lations
o)
Z
sos =
Zo) (1)
Z
9, y,
Zo) :
center
trix
h some initial
to see, whether
ge orientation is
s contain three
s presented in
ncters can be
the laser point
with the new
0 an iterative
d any more.
that there is no
is as well an
and and handle
o need to filter
in operator can
do not seem to
es even small
e used as a tie
,is to improve
1 color-coding.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Typically, the color-coding is done according the distance from
the camera location or according the altitude.
The interactive method requires enough visible features on the
image footprint. These features can be buildings, road signs,
fences and even trees, for example. Specially, if the close-range
images are used, the image footprint is usually quite limited and
may contain too few distinguishable targets for accurate
orientations. The panoramic images provide ultra-wide viewing
angle and therefore better ensures finding reliable set of features
within the image. Figure 3 is a part from panoramic image
mosaic, created from concentric image sequences. This method
of mosaicing is described in Haggrén et al. (1999) and Póntinen
(2000).
The interactive orientation method is applicable also for direct
relative orientation of two laser point clouds. Firstly, the first
point cloud is superimposed to the plain image plane, leading
the situation that actually equals to a normal central perspective
image. Secondly, the interactive orientation method is applied
to find relative orientation between this image of the first point
cloud and the other laser data set.
With synthetic images, there are no limitations for the
perspective of inspecting. Therefore, angle of view can be
chosen in a way the tie features are most visible. Typically, the
reference area should be investigated, at least, at two
perpendicular directions to ensure good accuracy in each
direction. In this research, the test sites were inspected from two
to six different angles of view. The described method to adjust
two laser point clouds directly into the common coordinate
system was applied the first time in Hyyppä et al. (2003) in the
forestry areas.
Figure 3. Laser scanning data provide good coverage of the
building. However, some small deformations are
detectable. This image covers about 23 % of the
original panoramic image and laser point cloud.
Comparison between laser strips was done in all thirty-nine
small test sites (Figure 2). The laser strip 2 was selected as a
reference strip and the other strips were oriented to that.
Because the test sites were quite small ones, only the shifts
between point clouds were solved. If there was any detectable
shift (e.g. in Figure 4), the difference was measured. Each
orientation was done independently, without knowing the
differences in surrounding test sites.
A) : TR a B)
A
4
. mr
J
eM
NN ee
i S
Figure 4. The planimetric shift of the building between laser
strips 2 (black) and 5 (white) is visible from two
different central perspectives. A) The shape of the
roof. B) The wall and the edge of the roof.
4. RESULTS
During the orientation process, it became obvious that the
differences should be presented in the along-track, across-track
and height direction. This is primarily, because the gaps in
scanning geometry (Figure 1) caused problems in many test
sites for orientations in across-track direction. In this research,
the corrections were measured only, if some differences were
detectable. Therefore, the distinct shift could be easily
underestimated, if it was not possible to improve orientation
due the scanner properties. To reduce this problem, some of the
Worst test sites were discarded from the across-track direction.
The results are presented in Tables I, 2 and 3.
Strip 2-3 | Strip 2-4 | Strip 2-5 | Strip 2-6
Mean [m] 0.050 -0.005 0.064 -0.010
Std [m] 0.039 0.018 0.041 0.023
Max [m] 0.150 0.025 0.136 0.045
Min [m] -0.009 -0.041 -0.010 -0.078
Table 1. Differences in flight direction (39 samples per strip)
Strip 2-3 | Strip 2-4 | Strip 2-5 | Strip 2-6
Mean [m] -0.012 0.003 -0.019 -0.005
Std [m] 0.027 0.015 0.034 0.019
Max [m] 0.018 0.037 0.025 0.035
Min [m] -0.085 -0.020 -0.099 -0.039
Table 2. Differences in across-track direction (20 samples per
strip)
Strip 2-3 | Strip 2-4 | Strip 2-5 | Strip 2-6
Mean [m] 0.001 -0.003 -0.002 -0.014
Std [m] 0.011 0.008 0.011 0.011
Max [m] 0.027 0.014 0.025 0.002
Min [m] -0.025 -0.027 -0.022 -0.043
Table 3. Differences in elevations (39 samples per strip)
The flight direction of the strips affects remarkably in the
obtained planimetric errors both in along- and across-track
directions. However, such phenomenon is not visible from the
heights. If the differences between strips 3 and 5 are examined,
the bias of only -0.014 m and standard deviation of 0.032 m in
flight direction can be found. Correspondingly, the bias in
across-track direction is 0.006 m with standard deviation of
0.027 m.
