ection), 8
; problem
run time
se look on
emory on
should be
e relations
roads and
vicinity e.
ions helps
> instances
f all lines.
1 structural
rmore one
his field in
is given to
be parallel
yy Fig. 3a).
ering. Then
So they can
/ drawing it
ck all lines
sketched in
nuch when it
on with data
ns that in our
assessed. We
Eckart Michaelsen
compared e. g. our system to one other shell and approach, that we know well enough to dare so. The current backlash
in the field of structural recognition methods may be explainable to some degree by the difficulties mentioned. But for
some tasks, there may be no other way out.
Complex models with many degrees of freedom permit classical one-step template matching (with tolerable effort) only
if there is enough prior knowledge (e. g. from maps). Otherwise perceptual grouping of object parts seems to be the
only reasonable alternative. This gives best results if an alternation is implemented between grouping and matching.
Generic models pose the most severe performance problems because of their inherent combinatorics. We discussed this
with an example of grouping buildings and roads into settlement structures. But computational complexity is only one
of the difficulties encountered. There are some other serious problems of which the applicator should be aware, if he or
she is confronted with a task, that appeals for such a method. Our advice is to check the list of problems with respect to
the special task, before one decides to use one of the alternatives mentioned in Sec. 4.1 or something else.
We do not want to discourage anybody. Perceptual grouping and model based recognition remains a fascinating and
promising discipline inside computer vision. The more problems you encounter the more ambitious becomes the topic,
and awareness of the problems helps to find clever solutions.
REFERNENCES
Ade F (1997) The role of artificial intelligence in the reconstruction of man-made objects from aerial images. In: Gruen
A, Baltsavias EP, Henricsson O (eds) Automatic extraction of man-made objects from aerial and space images (II).
Basel: Birkháuser
Knuth DE (1973) The Art of Programming. Vol. 3:Sorting and searching, Reading Mass.: Addison-Wesley
Kohonen T (1985) Content addressable memories. Springer, Berlin.
Kummert F, Niemann H, Prechtel R, Sagerer G (1993) Control and explanation in a signal understanding environment.
Signal processing, 3: 111-145
Matsuyama T, Hwang V (1990) SIGMA: A knowledge-based aerial image understanding system. New York: Plenum
Press
Metzger W (1975) Gesetze des Sehens: Die Lehre vom Sehen der Formen und Dinge des Raumes und der Bewegung.
Frankfurt: Waldemar Kramer
Michaelsen E, Stilla U (1998) Remarks on the Notation of Coordinate Grammars. In: Advances in Pattern Recognition,
Joint IAPR Workshops SSPR'98 and SPR'98 in Sydney, Berlin: Springer, 421-428.
Quint F (1996) Recognition of Structured Objects in Monocular Aerial Images Using Context Information. In: Leberl F,
Kalliany R, Gruber M (eds) Mapping buildings, roads and other man-made structures from images, IAPR-TC7. Wien:
Oldenburg, 213-228
Sulla U (1995) Map-aided structural analysis of aerial images. ISPRS Journal of Photogrammetry and Remote Sensing,
50(4): 3-10
Stilla U, Michaelsen E, Lütjen K (1996) Automatic extraction of buildings from aerial images. In: Leberl F, Kalliany R,
Gruber M (eds) Mapping buildings, roads and other man-made structures from images, IAPR-TC7. Wien: Oldenburg,
229-244
Stilla U, Michaelsen E (1997) Semantic modelling of man-made objects by production nets. In: Gruen A, Baltsavias EP,
Henricsson O (eds) Automatic extraction of man-made objects from aerial and space images (II). Basel: Birkhaeuser,
43-52
Wertheimer M (1912) Experimental Studies on the seeing of motion. Reprinted in: Shipley T (1961) Classics in
psychology. New York: Philosophical Library
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 583