Istanbul 2004
ındant because
lective area of
es initially the
of the target.
the target and
otations of the
s transferred to
ons are applied
1e XY-plane of
cm centered at
only the data
e classification
y the class with
estimating the
) correspond to
unds the centre
Ns cluster and
yrdinates of the
ims may seem
is significantly
is confirmed by
n.
using grid and
oridding. Using
| for the target
cloud. In order
y processed for
is fitted on the
of the distance
d, but only the
fidence) from
he grid is set to
ated. Using the
ated. Using the
st one is named
n the same way
ormation of the
77ygridrad” and
f the grid. The
| on delaunay
ed ‘delrad’ and
/ious ones is the
LTS
evaluate the
pare it with the
n literature. In
designed and
' several targets
EDM internal
s of the targets
| series involved
| on a wall. The
us distances and
iere subdued to
nal and external
aforementioned
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
For the first series of experiments, four targets mounted on the
pillars of the EDM calibration baseline were scanned. The
targets were placed at various distances that ranged from 3m to
25m. The scans of the same targets were collected from two
different positions, A and B. At position A, four scans of Imm
spacing were acquired along with a fine scan for each one of the
targets. At position B, nine scans of 1mm spacing and a fine
scan for each one of the targets were collected.
For the A position, the centres of the targets were calculated
using the fine scans, a single and four merged scans. For the B
position, the centres of the targets were also calculated for nine
merged scans. In both cases, using the coordinates of the targets
as derived from the fine scans as reference, and the coordinates
of the targets that were calculated for the other datasets of the
same position, the transformations were calculated. Also, the
mean absolute error was derived in order to evaluate the
internal accuracy of the algorithms. The results are summarized
in Table 3. Clearly, the performance of the ‘fuzzyposfine’
algorithm is superior.
This is also confirmed by the results presented in Table 4.
These results were derived using data from the second series of
experiments. Five targets that were mounted on a wall were
scanned 10 times each from a distance of approximately 5m
with the scanner facing directly the targets. For each one of the
targets a broad area containing the target was scanned. In order
to create the reference dataset, a single scan for each one of the
targets was used. For the reference dataset, the data were
trimmed so as to contain only the target. The other data were
exported as collected (along with the area that was surrounding
the target). Three datasets were created using a single, four
merged and nine merged scans for each one of the targets. The
*fuzzypos' and 'fuzzyposfine' algorithms once again perform
better, indicating that these algorithms have a very high internal
accuracy. The results of the ‘fuzzygridrad’ and ‘fuzzydelrad’
algorithms are also quite satisfactory in both cases. ;
The second part of the results refers to the evaluation of the
external accuracy of the algorithms. In the results to be
presented, both single and multiple scans collected from
different positions of the scanner are used.
For the case of the targets of the EDM baseline, the process of
calculating the mean absolute error was carried out using the
fine scan, a single scan and four merged scans from positions À
and B. Additionally, using the fine scans, the centres of the
targets were determined using the Cyclone software and the
registration process was carried out for the data that were
selected from the two positions of the scanner. The mean
absolute error of the transformation as derived by the Cyclone
software was Imm. This value is used later on for the
evaluation of the algorithms.
*
In Table 5, the results for the evaluation of the external
accuracy of the algorithms are presented. In this case, using the
fine scans, only the 'fuzzyposfine' algorithm gives a Mean
Absolute Error equal to that of the Cyclone software.
Additionally, using the other datasets, the results vielded by this
algorithm are better than Imm (i.e. 0.7mm for single scan
datasets and 0.8mm for datasets of four merged scans).
Table 3: Internal accuracy evaluation experiment (1)
Mean Absolute Error (mm) Mean
DATA Position A Position B Error
Ise d: 4dsc. 1.120. 1 disci. 9sc (mm)
radcent 1:2 I2 0.8 0.6 0.6 0.9
maxrad 168.174 | 132 | 92 9.4 13.2
maxrad4 14.0 1.34.51] 13:71] H3: E 1477 14.0
fuzzypos 0.7 0.7 0.6 0.4 0.4 0.6
fuzzyposfine| 0.6 0.6 0.5 0.2 0.1 0.4
gridrad 1.0 1.4 0.4 0.9 1.0 0.9
delrad 0.4 1.0 1] 1.0 1.6 1.0
fuzzygridrad| 08 | 0.6 | 0.7 0.5 0.7 0.7
fuzzydelrad | 0.9 12 1.1 1-7 1.8 1.3
Table 4: Internal accuracy evaluation experiment (2)
Mean Absolute Error (mm) Mean
reference data Error
I scan 4scans 9scans | (mm)
radcent 4.3 4.3 4.3 4.3
17.6 11:9. 8.2 12.6
DATA 10.4 8.4 8.9 9.2
2 0.2 0.2 0.2 0.2
T fuzzyposfine 0.2 0.1 0.1 0.1
= gridrad 4.7 4.7 4.7 4.7
delrad 3.8 3.2 4.1 3.7
fuzzygridrad 1.6 1.6 1.5 1.6
fuzzydelrad 1.4 1.3 1.3 1.3
Table 3: External accuracy evaluation experiment (1)
Mean Absolute Error (mm) Mean
DATA A fine A Iscan | A 4scans | Error
B fine | B Iscan | B 4scans | (mm)
radcent 1.4 2.4 2.4 2.1
maxrad 9.3 5.4 10.0 8.3
maxrad4 54 33 10.5 6.3
B bonmuypes |l 34 1.2 1.2 13
= fuzzyposfine 1.0 0.7 0.8 0.9
= gridrad 1.4 1.6 1.8 1.6
delrad ME 1.5 ES 1.4
fuzzygridrad 1.5 1] 1.2 13
fuzzydelrad 1.4 1.3 12 1.3
Table 6: External accuracy evaluation experiment (2)
Mean Absolute Error (mm) Mean
3m 10m Error
BATA 90°41 45° | 15° { 90°. 145% 15° | (mim)
radcent 42 | 49/64} 44 | 5.1 | 58 5.1
maxrad 13.01 14.4121.:0[25.4123.1] 19:6 1-198
maxradd {14.01 10.7 115.51 7.3 1 11.24185 | 12.9
S| fuzzypos [06109120907] 1.2 | 09
p fuzzyposfine| 0.4 | 0.6 | 07 | 0.4 | 04 | 04 | 055
= gridrad |148|52175[44[|52| 64 | 56
delrad 3.3.1.4.1.1 53.1.3543 AA 47.1 42
fuzzygridrad | 1.9 1.1.9. 12.9.| 1,61 1.8.].2.6 1. 2.1
fuzzydelrad | 2.0 | 1.8 | 2.7 | 1.4 | 1.4 | 2.3 1.9
