front right rear left
reference angle (degrees) 0 85 195 265
calibration result (degrees) 0 85,07 195,02 265,02
Table 1: Viewpoint calibration results
The registration procedure is illustrated in figs. 14 and 15. Begin-
ning with the front camera position a new viewpoint is registered.
The left image of fig. 14 shows the camera image of a new view-
point with the model's edges projected in the (yet) wrong camera
position. The projected edges do not match with the image edges.
Fig.14 (right) shows the model's geometry together with the ini-
tial front camera position in form of a pyramid.
Figure 14: Before registration of a new viewpoint
Figure 15: After registration of a new viewpoint
The situation after registration is depicted in fig. 15. The new po-
sition has been estimated leading to a projection of the geometry
that matches the image edges (fig. 15 left). The second pyramid
(fig. 15 right) depicts the estimated camera position.
An angle of about 45 degrees between two viewpoints could be
estimated without encountering numerical instabilities. In case
larger movements have to be estimated an initial guess had to
be made. The images of the reference object were taken with
one camera while the object was rotating in front of it on a com-
puter controled turntable. Due to the very accurate positioning
of the turntable the rotational angle between the images is ex-
actly known. Table 1 shows the calibration results for some se-
lected viewpoints. It should be stated that the presented estima-
tions were achieved only using a manually generated segmenta-
tion. By use of a subpixel-accurate edge detection the precision
could probably be further increased.
7 Conclusions
We have discussed the surface reconstruction of the system
AIDA, which extracts 3-D geometry from a sequence of stereo
images. It is particularly employable when reconstructing scenes
consisting of man-made objects like buildings. Through the use
of an explicit scene description in form of a generic semantic
net geometric constraints are selected during interpretation. They
restrict the model geometry to fulfill some expectations human
observers have about such scenes like rectangular, plane walls,
straight edges or symmetries.
This allows to further increase the model's quality compared to
the results that can be achieved by only using image data. In
particular properties that are important to human observers, like
symmetries and parallelisms, are improved. Especially the edge
constraint has proved to be very important since texturing de-
pends directly on the exact congruence between projected model
edges and image contours.
The approach presented in this contribution leads to realistic
models that can be employed e.g. for driving simulators or in
virtual scenes for film production.
References
[1] C. Braun, Th. Lang, F. Schickler, Steinhage, Cremers,
W. Fórstner, L. Plumer: On the Models for Photogram-
metric Building Reconstruction Computer & Graphics,
Vol. 19, No. 1, 1995
[2] O. Grau: Ein Szeneninterpretationssystem zur Mo-
dellierung dreidimensionaler Kórper, 17. DAGM-
Symposium Mustererkennung 1995, 13.-15. Sept. '95
[3] O. Grau: A Scene Analysis System for the Generation
of 3-D Models. Submitted Paper
[4] R.Koch: 3-D Surface Reconstruction from Stereoscopic
Image Sequences, International Conference of Compu-
ter Vision ICCV '95 Cambridge, MA., USA, June 1995
[5] F. Leberl, M. Gruber, P. Uray, F. Madritsch: Trade-Offs
in the Reconstruction and Rendering of 3-D Objects,
Mustererkennung 1994, 16. DAGM Symposium und
18. Workshop der OAGM, Wien 1994
[6] C.-E. Liedtke, O. Grau, S. Growe: Use of Explicit
Knowledge for the Reconstruction of 3-D Object Geo-
metry, 6th International Conference CAIP '95 Com-
puter Analysis of Images and Pattern. Sep. 6-8 1995
Prague
[7] Y. Yakimovski, R. Cunningham: A System for Extrac-
ting 3-D Measurements from a Stereo Pair of TV Ca-
meras, Computer Graphics and Image Processing Vol
7, pp. 195-210, 1978
[8] Pakzad, Kian: Segmentierung von Tiefenkarten auf-
grund ähnlicher Flächenorientierung, Studienarbeit an
der Universität Hannover, 1994
[9] S. Weik: Erzeugung dreidimensionaler Oberflächen-
modelle unter Einhaltung gemetrischer Randbedingun-
gen, Diplomarbeit Universität Hannover, Aug. 1995
[10] R. Tsai: A Versatile Camera Calibration Technique for
High-Accuracy 3-D Machine Vision Metrology Using
Off-the-Shelf TV Cameras and Lenses, IEEE Journal
of Robotics and Automation, Vol. RA-3, No.4, pp.323-
344, Aug.1987
598
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
KEY W
ABSTR
In this |
museum
in the se
KURZI
Im vor
Bildver:
photogr
von Me
Alle bec
welchen
interessi
archivie
môglich
für das
kommer
zusátzlic
gebiete
die vor
Unterscl
Hauptar
Topogra
immer p
beginnei
Meyden
und dis|
das im ‘
Gebiet d
In (Ack
digitaler
Wurde f
auf Film
war vor
Operatei
Interpret
Photogr:
Erkenne
kompleti
auszuwe
übertrag
auch In
gespeich
