level 0 (base level) level 1
Figure 4: triangulation levels for avalanche obstacles
level 2
level 3
Figure 5: triangulations for tree groups
5. Implementation
The topographic map at 1:25‘000 scale is composed of 6 sepa-
rately scanned layers of forests, forest boundaries, height
lines, buildings/streets, lakes and waters. Due to the size of
eachlayer (- 200*000 lines x 300'000 columns), an image cat-
alogue with an efficient tiling structure has been chosen for
the organisation of the data [Boesch 1995]. Each layer repre-
sents a binary image of the original map layer.
In a first step, a tile of an image layer is vectorised by la-
belling all connected components, allowing the calculation of
different shape descriptors (Figure 7).
Discrimination rule interpreters can lead to a serious reduction
in performance. Therefore, completed discrimination rules
(e.g. Table 1) have been implemented as dynamically loadable
libraries (often called plug-ins). Two or more plug-ins can be
grouped together to formulate abstract discrimination criteria
(e.g. the aggregate of all triangulation points and observation
towers) without considerable performance penalty.
The current implementation allows to process about IMPixels
per minute, depending on the complexity of the applied dis-
crimination rules.
Figure 6: triangulation for dry channels
AUTOMATIC
labeling and tracing
shape descriptors
shape discriminatio
triangulation
tree discrimination
Figure 7: process flow
INTERACTIVE
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
6. Conc
Shape des
powerful n
from scan
detect othe
to be pron
aerial phot:
single trees
Neverthele
approach h
* trying &
down a[
* using sl
inefficie
* discrimi
user inte
* detecta
lution
Referer
Bähr H.P,
L
Ji
2
Ballard D.
al
P
Ballard D.
P
Boesch R.
Ci
C
E
Chang Ji }
ti
N
Han Joon
{1
6
Henricsso
S
F
1
Kass M.,
n
C
L+T. „Poc
Q
[>
(
Lai Kok I
C
I
]
Lee H., R
