International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part-B2. Istanbul 2004
x 70120
70202
Row of
Trees Row of
Extracted Trees
Road
Extracted
Road
ë r Las J
Figure 4: Objects 70202 (left) and 70120 (right), Orthoim-
age and ATKIS ©Landesvermessungsamt Nordrhein-
Westfalen
relation and 4) and 5) the outcome from the combination.
The plausibility regarding the topologic relation is not de-
fined as —7' is not supported. On the z-axis the particu-
lar number of segments which have reached a certain sup-
port or plausibility value (10 bins on y-axis) is displayed.
The histogram-based analysis was chosen in order to ob-
tain a first overview on the quality representation. As all
observed ATKIS segments are supposed to be correct or
incorrect respectively, the histograms represent how good
the approach reflects the quality.
Three experiments have been accomplished with the cor-
rect data (Fig. 3): a) assessment of ATKIS segments using
extracted road objects, b) additional incorporation of rows
of trees and c) like b) but with decreased certainty and pre-
cision for ATKIS (set to zero). At first glance one might
be confused at the low support rates for the segments. This
can be explained by the relatively low number of extracted
objects: if at all, most ATKIS segments are not completely
covered by the extracted objects.
What is more interesting are changes from a) to c). Com-
pared to experiment a) the rows of trees (b) give a lot of
support and regarding geometry relations, the plausibility
also increases for most segments. The support regarding
topology increases mostly in the lower part (0.1 to 0.4) as
due to the relatively low certainties of the rows of trees the
support for this relation may not be very good in princi-
ple. It also corresponds to the model that in most cases the
support regarding topology in c) decreases as the quality
measures for ATKIS are set very strict“. In contrast the
support and plausibility regarding geometry just changes
marginally compared to b).
Two examples are chosen to clarify the sketched behavior
of the approach. The left image in Fig. 4 shows ATKIS ob-
ject 70202 being covered to approx. 50% by an extracted
road object and another 70% by a row of trees. The right
! If however the given geometry of an ATKIS segment coincides well
with the extracted object, the support from topology increases.
808
Exp... Z/Seg.. spT .spG-. pLG . sp.S. pls
a) 1 0,52 10.75 ; :0.:99: 03% 106
2 0.66" 0.71 097 0.90. 0.99
3 0.40" “039° 099 © 0.63" 100
b) 1 0.52 0.757099 0.858 100
2 0.66 (074 098-091 099
S 048 068 088 082 0.93
C) 1 0.13" 092 0.99: 094 20.99
2 007 082 097 033 097
3 034710778 70.92 0,82 095
Table 4: Results for 70202 (consists of 3 Segments)
Exp. :d/Ses. sp-To /sp.Gooc plGospS. «pbS
a) 657. 020 676 639. 035
D um m wo =
3 0.52 0.25 0797060 089
4 0:14 017-083: 027 — 0.85
b) 1 058 028° $73 7861 0%
2 0.00 039 100 039 1.00
3 054.035 "081 .00.67. 090
4 017 013 065 023 0.70
c) 1 0.25 026.070. 040 033
2 0.00 044 1.00 044 1.00
3 045 0.35 07% 060° 0:36
ES
0.04.0502 2074 6200144075
Table 5: Results for 70120 (consists of 4 Segments)
image show ATKIS object 70120 being covered to approx.
20% by extracted road and to another 20% by a row of
trees. The detailed assessment results for 70202 are given
in Tab. 4, Tab. 5 shows details for 70120. The first example
demonstrates very good support for the segments as these
are fully covered (compare experiment a) to b)). The sup-
port for the geometric relation in experiment c) increases
for some segments as the ROT decreases and therefore the
rows of trees in the northern part (which are not parallel to
the ATKIS segment) are not taken into account. In the se-
cond example the support is not very good as the coverage
is not sufficient. But the relatively high plausibilities al-
low the conclusion that the existing extracted objects give
not much evidence against the hypothesis that the ATKIS
segments coincide well with the model.
The experiments with incorrect ATKIS data (Fig. 5) were
conducted to investigate if the approach is able to detect
bad quality. Here two experiments have been carried out:
a) the extracted correct road objects are used to assess the
incorrect data and b) additionally the correct rows of trees
are incorporated. An outstanding property of both his-
tograms is that there is nearly no support beyond 0.3 for
both relations as well as for the combination. In some cases
in experiment (b) the plausibilities are relatively high. Such
situations may occur as similar to the support for the hy-
pothesis the support against it is not very high due to the
poor coverage.
6 CONCLUSIONS AND OUTLOOK
In this paper an approach for quality assessment of ATKIS
road vector data is introduced. A relationship model con-
tains the object class to be assessed and the object classes
in topologic and geometric relation to it. Besides the rela-
tions quality measures are defined having an impact on the
subsequent assessment phase. Here extracted objects are
assigned to the ATKIS segments according to the modeled
topologic relation and the respective given quality mea-
sures. Afterwards the existing topologic and geometric
Internati
ATKIS:
ATKIS:
F
relations
objects a
combine
Dempste
ATKIS s
plausibil
First res
data is r
work it
cision (v
jected) c
measures
ples that
tain thres
further re
Moreove
the simpl
oftopolo
does imp
Another
work asp
portant p
gives a c
sented ev
and cons;
