image quality, method of extraction, and 
line length). 
o The quality of interior and exterior 
orientation. 
e The imaging geometry. Mainly the angle between 
the planes is of importance for interpretation plane 
intersection. 
In the example of section 3 the angle between the interpretation 
planes of two corresponding horizontal lines is very small due 
to the fact that the line between the two projection centres is 
also horizontal and parallel to the building. Indeed the 
horizontal 3D lines show large errors (Figure 12). In 
conclusion, before intersection the angle between interpretation 
planes has to be checked for an acceptable minimum value. 
In order to avoid loosing many lines as in the example, rules of 
thumb have to be formulated for the image acquisition. For 
example, images taken at different height levels would improve 
significantly the intersections in section 3. Two other solutions 
can be chosen as well. Firstly, the 3D lines that result from bad 
intersections can be positioned in the a priori object plane when 
the precision of the plane position is better than the (depth) 
precision of the intersection. Secondly, the projections of 
horizontal and vertical edges of the building could be linked 
when their endpoints are only a few pixels apart, assuming that 
their intersection corresponds to a point on the building. Then 
the four related interpretation planes (two for the horizontal and 
two for the vertical edge) could be intersected using least- 
squares adjustment (van den Heuvel, 1999). More general, this 
adjustment involves a number of interpretation planes that 
equals the number of images (possibly more than two, see next 
section) times the number of linked edges. 
4.3 Multiple image matching 
In the current approach, the intersection of the two 
interpretation planes is not redundant. Reliability can be 
obtained by taking into consideration one or more additional 
interpretation planes utilising more images. In general, the 
quality of the intersected 3D line in terms of precision and 
reliability improves with each additional image in which the 
same object edge is extracted. Furthermore, the quality of the 
matching improves, i.e. the number of erroneous matches is 
expected to drop. Whilst in close-range photogrammetry 
considerable research efforts have been directed to multiple 
image point matching (Maas, 1992), multiple image line 
matching is still insufficiently explored. In this respect we 
consider the image line matching a promising topic for future 
research. 
5. CONCLUSIONS 
We have presented our preliminary results on 3D line extraction 
to provide line features needed for the accurate positioning for 
an augmented reality application. Although the approach was 
inspired by the 3D model reconstruction procedures applied 
within the UbiCom project, it contributes to the more general 
research on 3D line extraction. The obtained results exhibit a 
number of positive findings as the edge matching is concerned. 
Further research has to be carried out for improving the 
interpretation plane intersections to obtain the 3D line feature 
parameters. Utilisation of multiple images should be one of the 
first steps towards improvement. 
REFERENCES 
Foerstner, W., 1994, A framework for low level feature 
extraction, in: Computer Vision - ECCV94, Springer Verlag, 
Berlin, vol.2, pp.283-394 
van den Heuvel, F.A., 1998, Vanishing point detection for 
architectural photogrammetry, International Archives of 
Photogrammetry and Remote Sensing, Vol. XXXII part 5, 
Hakodate, Japan, pp.652-659 
van den Heuvel, F.A, 1999, A  Line-photogrammetric 
mathematical model for the reconstruction of polyhedral 
objects. in Videometrics VI, Sabry F. El-Hakim (ed.), 
Proceedings of SPIE Vol.3641, ISBN 0-8194-3112-5, pp. 60-71 
Maas, H.-G., 1992, Robust Automated Surface Reconstruction 
with Structured Light. in: International Archives of 
Photogrammetry and Remote Sensing, Vol.29, Part B5, 
pp.709-713. 
Pasman, W., and F. W. Jansen, 2001, Distributed Low-latency 
Rendering for Mobile AR, in: Proceedings of IEEE & ACM 
International Symposium on Augmented Reality, 29-31 October, 
Columbia University, NY, pp. 107-113. 
Pasman, W., S. Zlatanova, S. Persa, J. Caarls, 2001, 
Alternatives for optical tracking, UbiCom internal report 
30/5/1, http://www.cg.its.tudelft.nl/~wouter/publications/, 
(accessed June, 15^ 2002), 12p. 
Persa, S. and P.P. Jonker, 2000, Hybrid Tracking System for 
Outdoor Augmented Reality, in: Proceedings of 2nd Int. 
Symposium on Mobile Multimedia Systems and Applications 
MMSA2000, 9-10 November, Delft, The Netherlands, pp. 41- 
47. 
Suveg, I. and G. Vosselman, 2002, Automatic 3D 
Reconstruction of Buildings from Aerial Images, in: 
Proceedings of SPIE Photonics West, Electronic Imaging, 19- 
25 January, San Jose, USA, CDROM 
Ubicom project, 2002, http://bscw.ubicom.tudelft.nl/ (accessed 
June, 15™ 2002) 
Vermeij, M. and S. Zlatanova, 2001, Semi-automatic 3D 
building reconstruction using Softplotter, in: Proceedings of 
International Symposium on “Geodetic, Photogrammetric and 
Satellite technologies: development and integrated 
applications ", 8-9 November, Sofia, Bulgaria, pp. 305-314 
Zlatanova, S. 2001, 3D modelling for augmented reality, in: 
Proceedings of the 3" ISPRS Workshop on Dynamic and Multi- 
Dimensional GIS, 23-25 May, Bangkok, Thailand, pp. 415-420 
Zlatanova S. and F. van den Heuvel, 2001, 3D city modelling 
for mobile augmented reality, in: Proceedings of International 
Symposium on Architectural Photogrammetry, 8-21 September, 
Potsdam, Germany, 8p. (to be published). 
—233—