Calibration, point matches, and line segment matches
                             of a close-range architectural image set
                       ----------------------------------------------------

This set contains images of an Oxford University building, taken by
Olympus C-820L digital camera. The camera was held by a person
standing on the ground. The following was applied to the set:

1. Automated wide baseline stereo: affine & scale invariant features
were detected in each image and matched by RANSAC, yielding projective
camera matrices and a cloud of projective 3D points.

2. Metric upgrade from vanishing points: vanishing points computed
from detected line segments and (along with square pixels assumption)
used to upgrade the projective reconstruction to metric one.

3. Automated line segment matching: the line segments matched and
reconstructed, using geometrical constraints on lines and photommetric
information in their neighborhood.
   There are some mismatches among the matched lines, mainly due to
ill conditioning of the geometry for horizontal lines (because camera
centers all lie in about horizontal plane).


References:

1. Werner, T. and Zisserman, A.
   New Techniques for Automated Architecture Reconstruction from Photographs
   Proc. 7th European Conference on Computer Vision, Copenhagen, Denmark, 2002.
   http://www.robots.ox.ac.uk/~vgg/publications/papers/werner02.{pdf,ps.gz}.

2. Werner, T. and Zisserman, A.
   Model Selection for Automated Architectural Reconstruction from Multiple Views
   Proc. British Machine Vision Conference, 2002.
   http://www.robots.ox.ac.uk/~vgg/publications/papers/werner02a.{pdf,ps.gz}.