CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey
The basic idea of Zitnick and Kanade (2000) is a cooperation
between support and inhibition. Inhibition enforces the
uniqueness of a match. Assuming opaque and diffuse-
reflecting surfaces, a ray of view emanating from a camera will
hit the scene at one point only. The idea is to gradually weight
down all matches on a ray besides the strongest. Support is
realized by filtering the 3D array by a 3D box filter. By this
means all matching scores corroborate locally to generate a
continuous surface. We have improved this scheme with
different means, most notably a combination of image and
disparity gradient to avoid smoothing away details and a
detailed treatment of occlusions. Additionally, we use the
information of a third image by projecting the results of the
cooperative computation into the third image by means of the
trifocal tensor and by modification of the 3D array based on
the correlation score of the first and the third image.
Figure 7 gives the result for the computation of the disparity
map from three images of the image sequence given above.
The disparity map outlines the important structures of the
scene. While moving from left to right as done in the first two
visualizations results into the movement of the wall in front of
the camera and an attractive view, a movement upwards as in
the rightmost image shows problems from non-modeled
occlusions.
4.4 Analysis of the Automatic Approach
We have shown a way to fully automatically generate metric
3D structure from the weak information of the perspective
images of a sequence. Yet, there is ample room for
improvement before this approach will become practically
relevant. Most notably, it is necessary to link the results for
structure computation in two or three images, to obtain a
coherent result for the sequence. Then, there is a host of
robustness issues which has to be solved before all types of
sequences can be handled reliably. Finally, in the foreseeable
future, the automatic approach will have problems, when the
perspective skew between images is too large, i.e., when the
angle between consecutive images is too large. This can be
avoided by taking more images. This is disadvantageous when
taking the images, but it helps to avoid a lot of manual work.
5. CONCLUDING REMARKS
Image-based survey and 3D modeling of architectural objects
can be performed by photogrammetric as well as computer
vision and computer graphics methods. The latter allow fully
automatic feature extraction, orientation, and 3D object
reconstruction, even if the knowledge about the geometry of
the image sequence is weak. Auto-calibration is possible in
case that the interior orientation data have changed during the
image recording (focusing, zoom) or are not available at all.
Preferably, photogrammetric and computer vision methods are
integrated to generate virtual models from images as precise
and reliable as required.
It should be noted that the results presented in this paper are
preliminary. The reconstruction and modeling of Wartburg
Castle is still in progress.
6. REFERENCES
Daskalopoulos, A., Georgopoulos, A., Ioannidis, Ch., Makris,
G.N., 2003. A Trip from Object to Image and Back. The
International Archives of the Photogrammetry, Remote
Sensing and Spatial Information Sciences, Ancona / Italy, Vol.
XXXIV (5/W12), pp. 141-144.
El-Hakim, S.F., 2002. Semi-Automatic 3D Reconstruction of
Occluded and Unmarked Surfaces from Widely Separated
Views. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, Corfu /
Greece, Vol. XXXIV (5), pp. 143-148.
Fischler, M., Bolles, R., 1981. Random Sample Consensus: A
Paradigm for Model Fitting with Applications to Image
Analysis and Automated Cartography. Communications of the
ACM, 24 (6), pp. 381-395.
Freeh, I., Koch, B., 2003. Multimedia Geoinformation in Rural
Areas with Eco-Tourism: The ReGeo-System. In: Information
and Communication Technologies in Tourism 2003, Springer,
Vienna / New York, pp. 421 -429.
Hanke, K., Oberschneider, M., 2002. The Medieval Fortress
Kufstein, Austria - An Example for the Restitution and
Visualization of Cultural Heritage. The International Archives
of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, Corfu / Greece, Vol. XXXIV (5), pp.
530-533.
Hao, X., Mayer, H., 2003. Orientation and Auto-Calibration of
Image Triplets and Sequences. The International Archives of
the Photogrammetry, Remote Sensing and Spatial Information
Sciences, Munich / Germany, Vol. XXXIV (3/W8).
Hartley, R., Zisserman, A., 2000. Multiple View Geometry in
Computer Vision. Cambridge University Press, Cambridge,
UK.
Mayer, H., 2003. Analysis of Means to Improve Cooperative
Disparity Estimation. The International Archives of the
Photogrammetry, Remote Sensing and Spatial Information
Sciences, Munich / Germany, Vol. XXXIV (3/W8).
Pollefeys, M., Verbiest, F., van Gool, L., 2002. Surviving
Dominant Planes in Uncalibrated Structure and Motion
Recovery. In; Seventh European Conference on Computer
Vision, Vol. II, pp. 837-851.
Pollefeys, M., van Gool, L., Vergauwen, M., Cornelis, K.,
Verbiest, F., Tops, J., 2003. 3D Capture of Archaeology and
Architecture with a Hand-Held Camera. The International
Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, Ancona / Italy, Vol. XXXIV (5/W12),
pp. 262-267.
Zitnick, C., Kanade, T., 2000. A Cooperative Algorithm for
Stereo Matching and Occlusion Detection. IEEE Transactions
on Pattern Analysis and Machine Intelligence, 22 (7), pp. 675-
684.
108
