was not measured, which makes the comparison to other
measurements slightly biased. ( VM )
University of Stuttgart ( TUS )
The measurements on digital imagery were made using
visual and automated methods. In the automated
measurements, the special features were automatic
selection of tie points, feature based matching and high
number of tie points. A basic difference, as compared to
other methods, is also the way tie points are used. The
points are in small patches, which correspond to single
points in the other methods. Very typically, the points
were measured on only two images.
The Zeiss PK1 measurements were made in a
conventional way for comparison.
TUS 1-4 Zeiss PK1 monocomparator measurements
using analogue imagery. The cases have
often been referred to with the name PK1.
(VM)
TUS5-12 Visual observations on digital images.
(VM)
TUS 13-16 Automatic measurements using 15 um data
( 272 points per image ). ( AM)
TUS 17-20 Automatic measurements using 30 um data
( 237 points per image ). (AM)
TUS21-24 Automatic measurements using 30 um data
( 131 points per image ). ( AM)
Technische Universitat Wien ( TUW )
The method is based on feature based matching, but, in
this case, unlike TUS, the number of points corresponds
to conventional cases. ( SM)
Université Laval ( UL )
The measurements were made completely visually using
Leica DVP system. The measuring resolution of 1/2 pixel
is lower than in most systems. ( VM )
5. ANALYSIS OF THE RESULTS
The analysis of accuracy is based on three methods;
e Calculation of root mean square error values ( RMS )
from differences between known and adjusted check
point coordinates.
e Calculation of RMS values from short horizontal
distances of neighbouring traverse points.
e Theoretical error estimates delivered by the block
adjustments.
Most of the analysed cases are based on dense control
and additional parameters.
5.1 15 um pixel imagery
The results with 15 um pixel size, dense control, and
additional parameters are presented in Figure 3.
352
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
In plane coordinates, the RMS values, the variations of
which are relatively small, are around 20 mm. An
explanation for bigger values in TUM 2 is the one pixel
measuring resolution, and in TUW 2 the low number of
points when using feature based matching.
80
70
60
50
mm 40
30
20
10
0
mium umemem TEEN. MWR IZ 'Zz'z 0190040 CDD
9:30000000E FE À Sc S GS HD XX
mem: —0xpb mu IIOÓrTIZzz002m..-
T: m2 =D IN oN o N i
e nN oO 0o OQ» O C N No o » M
Figure 3. RMS values from 15 um pixel size data, as
well as from BC1 and PK1 measurements
when using dense control and additional
parameters.
In comparison to measurements using analogue imagery,
the best results are at the same level with the BC1
measurements, and only slightly worse than the PK1
results. The used block with signalised check points is
favourable for visual methods, in general, and especially
for monocomparator measurements.
In a comparison between the different measurements of a
single participant, no big differences can be found. An
increase of measuring efforts, like in the series of FGI,
seems to have no influence in planimetric accuracy.
In the Z-coordinate, the variation is larger. The largest
errors have often occurred, when using visual methods
( FGI 2, NLSS 2, NTH 2, TUM 2, TUM 6, TUS 6 and
UL 2 ). An exception from this is NLH 2.
25
20. n Ll. n. igi
45 E 5 :: ET A E A i à E 5 E: E: | m
| | L| | | »m" B i E| E.
mm 1 HOH BH ET 3 EH B EB
DLL uM M M M- Hd 1H BH. Nm B B BI
nmnaeadds2ogoBoszzasdssdemns
m EOD UU Niele I I 0 1 2 2 0 0 SNA
2M "Moo! o à 9 N o NN O ES
Estimated local coordinate errors
calculated from short distances, when 15
um imagery, dense control, and additional
parameters have been used.
The coordinate accuracies calculated from short
distances ( Figure 4 ) show that the pointing accuracy on
check points has varied. These values have an effect on
values in Figure 3.
5.2 30 pm pixel imagery
The results from 30 um data with dense control and
additional parameters ( Figure 5) have somehow a
different nature than the corresponding results using 15
um data. The variations of RMS values between different
cases are, in general, larger, and the height-to-planimetri-
accuracy relation is in most cases clearly smaller.
100
80
mm 60
40
20
Figure
The im
from ca
a subs
the 20
adjustn
observ:
TUS 18
An ano
the cas
measur
53 E
The e
compar
sizes. |
plane, :
remark:
feature
which i:
mm 4
NY
Figure
In heigl
In mos
using ©
as for :
An ex
larger
visual «
the bac
on a fle
