method of the
same direction
the opposite
cantly smaller
slopes (i.e. to
ted for every
it strips. The
mode derived
h building and
c.f. Table 8.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Direction 2F- 3F- 4F- SF- 6F-
of edge RTK RTK RTK RTK RTK
ACTOSS 34x13 21:16 18:49 | -30+38 217
Along 77 15:43 14x11 24412 13:9
Oblique 19:46 31331 34+25 27328 | 21:49
Total 20:37 27426 26:3] 13436 19:38
Mean
difference
11.6 96
Table 9. Mean differences (cm) and standard deviations of the
roof lengths from first pulse mode observations and
RTK measurements.
The mean difference of the roof lengths across the flight strip
number 5 (first pulse) is negative. In this case, the laser
observations from the edge of some roofs were not accepted to
the roof plane because their undulations were too big. Hence
the roofs became shorter than in reality. The along flight strip
results for the first pulse mode observations (7-24 cm) are
expected to be smaller than the across flight strip results (18-34
cm) because of the scan pattern. The standard deviations were
quite small, between £3-12 cm. In total the roof lengths from
the first pulse mode are about 21 cm longer than the reference
lengths from RTK measurements.
14.0 %
Direction 2L- 3L- 4] ~ SL- 6L-
ofedge RIK RIK RTK RTK RTK
ACTOSS -52+30 | -40+11 -26x8 -81+13 -29+5
Along =35132 | 432 | 2721 | -32422 | 4723
Oblique -18+31 -9+39 -7+23 -12+16 | -1328
4.4 %
Total -29433 | -21x36 | -19+23 | -32+33 | -23425
13.2 %
11.4%
the last pulse
latively small
ulse measured
(4.4 %) when
s flat or pitch
at were lower
> observations
systematically
neasured roof
with the laser
lengths were
anes. Because
rn the lengths
1e flight strip,
. The distance
ction is about
> flying height
»verlap on the
Table 10. Mean differences (cm) and standard deviations of the
roof lengths from last pulse mode observations and
RTK measurements.
The last pulse mode results in Table 10 for the along track
direction do agree with the relation depicted in Table 9 between
across and along track directions. Again the across track value
(-81 em) of the flight strip number 5 indicates the shortest roof
lengths. In total the roof lengths from the last pulse mode are
about 25 cm smaller than the reference lengths from RTK
measurements.
4. DISCUSSION
The obtained elevation accuracy for repeated strips was good.
Also planimetric errors were mainly detected by comparing
strips flown with opposite directions.
The mean height errors for elevation points were —2 to +1 cm
and standard deviations were mainly +3-4 cm and for one strip
it was £9 cm. These values are comparable to other previous
results for planar targets (e.g. Ahokas et al., 2003).
Centre points and ridges of the extracted buildings were used to
test how small planimetric changes in the laser data affect the
obtained building model. Therefore, we also indirectly analyzed
the accuracy of the building extraction model. In paper by
Rónnholm (2004), a similar description of the systematic
internal quality of repeated measurements using point cloud
data is given.
The mean distances between the centre points of the buildings
derived from the first and last pulse observations differed less
than 20 cm from each other while the standard deviation was
+9-16 cm (Table 6). In one flight strip these errors include the
building extraction modelling errors and the errors resulting
from the differences between the first and the last pulse data.
Normally the shape of the building was the same, only area was
changed. When we compared the mean distances between the
centre points of the laser derived buildings and buildings on the
map, less than 30 cm differences occur. Standard deviations
were £11-28 cm and +14-18 cm (Tables 4 and 5). These mean
errors include the building extraction modelling errors, the
systematic shift between the transformed map and laser
coordinates and also the along track shift of the laser
observations to the fight direction. This gives information how
well the buildings can be extracted in real life from laser
scanner data. Vógtle et al. (2000) reported that the interior
accuracy of the modeled buildings is about 420-30 cm in
position and £5-10 cm in elevation. When they compared the
wireframe model with the manually derived CAD model the
coordinate differences were about +20-90 cm in xy and about
+20 cm in z. Steinle et al. (2000) reported the differences
between the first and last pulse derived wireframe models. In
this case the first and last pulse flights were from different dates
and the grid size was 1x1 m?. Mean differences were about 0.8-
1.5 m in xy and about 0.2-0.6 m in elevation.
First and last pulse data give different results for building sizes.
The total mean differences of roof lengths from first and last
pulse varied from about —30 to +30 cm and the standard
deviation was less than +40 cm (Tables 9 and 10). Buildings
were of different size and orientation. Some errors are by
inheritance from the observation geometry of the Toposys ALS.
An internal precision of 10-20 cm for dimensions of 10
buildings was reached in a study of Maas et al. (1999). The
FLI-MAP system produced a point density of 5 points/m°.
The results showed that the same ridge can be extracted by
repeated measurements with less than 3-5 cm systematic error
and the standard deviation of the shifts between acquisition was
less than 5 cm when using strips flown in the same direction. In
general this implies that the ridges extraction is very accurate
and the repeatability of the laser scanning with Toposys Falcon
is good.
5. CONCLUSIONS
The repeated observations of five Toposys Falcon flight strips
from Espoonlahti area were compared with each other and with
reference data.
Using reference measurements, the systematic height errors for
elevation points were —2 to +1 cm and standard deviations were
mainly +3-4 em. It can be concluded that the changes in height
errors of these five strips is negligible.
In planimetry, mean distances of the centre points of the
buildings were less than 30 cm for the first and also for the last
pulse data when compared with the buildings on the map. The
standard deviations varied between £11-28 cm (first pulse) and
+14-18 cm (last pulse) in different flight strips. The planimetric
accuracy of the object (building) on the ground depends on the
direction of the flight. There is an along track shift of about 5
cm to the flight direction. The accuracy obtained using the
centre points of the buildings depends on the extraction
accuracies and the distribution of the laser points on the ground.
Results obtained using the ridges confirmed that there is a
difference in planimetric accuracy between the flight strips to
the opposite direction and to the same direction. Biases were 3-
5 cm and standard deviations +3-5 em for the flight strips to the