'oductions
ing instances, two
| 2x5 windows and
ect. The upper row
(one row of eight
But there are also
size on the middle
hase of the middle
ced left and a little
ig is found.
‚USION
d. But we can state
part-of structure —
ifferent ways using
between. Here we
of façade objects
t while the systems
>ft direction — eg.
ot yield the same
recognition. In fact
nse decomposition
- each row consists
ist of an upper and
uch U-structure is
won't work for
ystem in reducing
ig real data — in
f additional clutter
1 as from thermal
rimitives must be
; should be tested
h as grouping the
ng the windows
à row.
e reliability of the
missing detections
e algorithm. Poor
re 4 and Figure 5)
iques, which sticks
this process small
; Figure 3 also à
ognize windows.
's would reduce
nts given in the
contain threshold
jar with the issue.
The systems compared here use of course the same constraints —
where possible. But some predicates do appear only in one or
the other variant, and an unskilled setting in one variant might
lead to an unfair comparison. This can be fixed when all such
parameters are optimally chosen based on sufficient and
representative data labelled by experts.
Minor further dependence of the results may be seen in different
parameters used in the interpretation search — e.g. the number of
parallel threads, overall time-limit, or top-down settings. The
latter is switched off here, in order to improve comparability.
And the computational effort was chosen large enough so that
little influence can be assumed. Experience shows that also the
setting of the parameters of the decision step is of little
influence to the result.
4.1 Outlook
More experiments are needed, in particular also regarding row
(and lattice) grouping according to the constant double ratio
principle of pinhole projection. This could be performed on the
original images, avoiding any re-sampling. Hopefully the
displacement problems are not so bad in that case.
Missing detections might be treated by extrapolation. That is by
prolonging the best gestalts and thus generating hypotheses
about the position and sizes of the windows with high precision.
Then an appearance model can be averaged from the gray
values found at the known positions and matched with the
values found at hypothesis locations.
The a vertical constraint demanding that window columns
should also be vertically grouped may be added, fostering
acceptable results on difficult data, such as here in the middle
row. And the interpreter is too slow. There should be ways of
improving it by hash techniques etc.
References
Ali, H., Seifert, C., Jindal, N., Paletta, L., and Paar, G. (2007)
Window detection in facades, 14th International Conference on
Image Analysis and Processing (ICIAP), vol. 1, pp. 837-842.
Ali, H., Ahmed, B., and Paar, G., (2008) Robust window
detection from 3d laser scanner data. Congress on Image and
Signal processing (CISP '08), vol. 2, pp. 115-118.
Burochin, J., Tournaire, O., and Nicolas, P. (2009) An
unsupervised hierarchical segmentation of a facade building
image in elementary 2d models. ISPRS Workshop on Object
Extraction for 3D City Models, Road Databases and Traffic
Monitoring - Concepts, Algorithms and Evaluation (CMRT 09).
Foerstner, W., Neumann, B., Sara, R., Petrou, M., Hotz, L.
(2009): eTRIMS - E-Training for Interpreting Images of Man-
Made Scenes. http://www.ipb.uni-bonn.de/projects/
von Hansen, W., Michaelsen, E., Thoennessen, U. (2006):
Cluster analysis and priority sorting in huge point clouds for
building reconstruction. In: Tang, Y.Y. (ed.): ICPR 2006, vol.
l., pp. 23-26.
Hoegner L, Stilla U (2007) Automated generation of 3d points
and building textures from infrared image sequences with ray
casting. In: Stilla U, Meyer H, Rottensteiner F, Heipke C, Hinz
S (eds) PIAO7 - Photogrammetric Image Analysis 2007.
International Archives of Photogrammetry, Remote Sensing and
Spatial Geoinformation Sciences, Vol 36(3/W49B):65-70
Hoegner L, Stilla U (2009) Thermal leakage detection on
building facades using infrared textures generated by mobile
mapping. Joint Urban Remote Sensing Event (JURSE 2009).
IEEE
Iwaszczuk D., Hoegner L., Stilla U. (2011) Detection of
windows in IR building textures using masked correlation. In:
Stilla U, Rottensteiner F, Mayer H, Jutzi B, Butenuth M (Eds.)
Photogrammetric Image Analysis, ISPRS Conference -
Proceedings. Lecture Notes in Computer Science, Vol. 6952,
Springer: 133-146
Lee, S. and Nevatia, R. (2004) Extraction and integration if
window in a 3d building model from ground view images.
Proceedings of the 2004 IEEE Computer Society Conference on
Computer Vision and Pattern Recognition, vol. 2, no. 1, pp.
113-120.
Liu, Y., Rauschert, I. (2011): Symmetry Detection from Real
World Images. Competition and workshop along with the
CVPR
http://vision.cse.psu.edu/rescarch/symmComp/index.shtml
Matsuyama, T., Hwang, V.S.-S., (1990) Sigma a Knowledge-
based Image Understanding System. Plenum Press, New York.
Mayer, H. and Reznik, S. (2006) MCMC Linked with Implicit
Shape Models and Plane Sweeping for 3D Building Facade
Interpretation in Image Sequences. In: International Archives of
the Photogrammetry, Remote Sensing and Spatial Information
Sciences, Vol. (36) 3, pp. 130-135.
Mayer, H. and Reznik, S. (2008) Implicit Shape Models, Self-
Diagnosis, and Model Selection for 3D Facade Interpretation.
Photogrammetrie - Fernerkundung - Geoinformation 2008 (3),
S. 187-196
Michaelsen E., Doktorski, L., Litjen, K., 2011. An
accumulating interpreter for cognitive vision production
systems. Pattern Recognition and Image Analysis, 21 (3), pp.
410-414.
Michaelsen, E., Stilla, U., Soergel, U., Doktorski, L., 2010.
Extraction of building polygons from SAR images. Grouping
and decision-level in the GESTALT system. Pattern
recognition letters, 31 (10), pp. 1071-1076.
Pu, W. and Vosselman, G. (2009) Refining building facade
models with images. ISPRS Workshop on Object Extraction for
3D City Models, Road Databases and Traffic Monitoring -
Concepts, Algorithms and Evaluation (CMRT'09)..
Ripperda, N. (2008) Grammar Based Facade Reconstruction
using RjMCMC. PFG Photogrammetrie Fernerkundung
Geoinformation. Stuttgart: Schweizerbartsche
Verlagsbuchhandlung, vol. 2008(2) pp. 83—92
Sirmacek, B. (2011a) Graph Theory and Mean Shift
Segmentation Based Classification of Building Facades. Joint
Urban Remote Sensing Event (JURSE'll), Muenchen,
Germany.
Sirmacek, B., Hoegner, L., and Stilla, U. (2011b) Detection of
windows and doors from thermal images by grouping
