d by im-
as it only
ified pa-
| images
tion. Us-
0 images
ric block
ency and
latic pro-
able.
duction,
emiauto-
ollowing.
high con-
e. They
ion is in
ig on the
| produce
:ecial in-
idows on
time the
o limited
machine
is recov-
selecting
rocedure
s at eave
provided
natching
natching
lel which
ixing the
gory the
internal
s for the
therefore
sponden-
matically
fferences
; optimal
pose or
order to
teraction
> missing
from the
ns, eases
by non-
ould use
f several
ple over-
| sides of
the building, being useful for either getting full information on
the building, not omitting occluded details or for orthophoto
production, which definitely requires to have information on
all surfaces visible from zenith.
Exploiting Tools from Digital Cartography. Interactive
systems for data acquisition have been developped since sev-
eral decades. All tools available there should also be used
here to advantage for easing interaction.
The snap function mentioned above is an example. The sys-
tem automatically checks a set of relations between different
primitives within a CSG-tree. The definition of that set of
course is context dependent. Here we started in looking for
common faces of primitives to glue them together, if the user
confirms the relation.
3 EXPERIMENTAL RESULTS
The system has been tested on large data sets. The perfor-
mance and accuracy of the system is investigated showing
promising characteristics.
3.1 Test field München
The test field München contains appr. 2 km? of the central
part around the area of the university, with large but complex
buildings. The data specifying the test site are given in the
table. The acquisition times are given in table for the three
generalization levels, adapted from [Lôcherbach, 1995].
image scale | 1: 15 000
focal length 153 mm
imagery B/W
pixel size 15 um
area 2 km?
# of models 578
# of prisms 230
Table 3: Information on test area München
generalization | models | total time | time/model
level [min] [min]
2 (medium) 249 461 1.85
3 (high) 295 548 1.85
4 (very high) 34 67 1.97
prisms time/prism
2 (medium) 7 18 2:57
3 (high) 128 446 3.48
4 (very high) 95 284 2.99
Table 4: Acquisition times for the test area München,
adapted from [Lôcherbach, 1995]
3.2 Test field Oedekoven
The testfield Oedekoven, near to Bonn, has been selected to
acquire data for the first phase of an OEEPE-Test on 3D-City
Information (cf. [Gülch, 1996]. The area covers about 1.5
km?, partly is sloped and contains buildings of many types.
The statistics on the test data are given in the table.
265
image scale 1: 12 000
focal length 150 mm
imagery B/W
pixel size 11 um
area 1.5 km?
# of CSG-trees 672
# of primitives 1591
# of primitives
per CSG-tree 2.5
% of boxes 66 %
% of gabled roof 18 %
% of hip-ed roof 3%
% lop-sided gabled roof 13 %
Table 5: Information on test area Oedekoven
Figure 4: Netto time per primitive with median and 25 %-
and 75 %-point,
250 T T T T T T T
2 "ebsl.plt" 9
o "ebsi.plt.25* —-
"ebnl.plt.50" -----
"ebsl.pit.75* —-—
200 d
9 o
; 8
150 + 8 7
o
i $
: di
L © 4
100 1 ° ? o ?
a °
$ dE e
- 26 03. mes a
ah ae En Ee d
> a 1 $ " 2 ME i-
= $ 6
a e 9
$ o
0 $ L 1 1 L L
0 1 2 3 4 5 6 7 8
The times for acquiring the data are given in the following
table. They contain all parts of the acquisition, including
navigation through the image pyramid, selection of primi-
tives, form adaptation, measurement of homologous points
and 3D-visualization. We distinguished 4 levels of generaliza-
tion: low generalization (1) corresponds to the finest details
identifiable, medium level of generalization (2) and higher
level correspond to leaving out details of 0.5 m, 1 m and 1.5
m appr.. The numbers for the different generalization levels
in the table do not correspond to the same area.
generalization models, total | brutto time/
level primitives | time model
[min] [min]
1 (low) 1591 | 2427 1.53
a) 1007 | 1181 1.17
b) 584 1246 2.13
Table 6: Acquisition times for the test area Oedekoven
a) expert, mainly flat terrain
b) non-expert, partly sloped terrain
We distinguish brutto and netto time, the difference being the
time for navigation, the time for deciding on modelling the
structure of the building and the time for checking the result.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
