The Béguinages of Leuven

Having university buildings on the UNESCO World Heritage list, we couldn't resist applying our 3D modeling techniques to it. In Figure 9.6 a view of the Béguinages of Leuven is given. Narrow streets are not very easy to model.

Figure 9.6: View of the Béguinages of Leuven

Using the presented technique we were able to reconstruct 3D models from video sequences acquired with a digital video camera. This was only made possible through the use of the polar rectification since the epipoles were always located in the image. An example of a rectified image pair is given in Figure 9.7. Note that the top part of the rectified images correspond to the epipole.

Figure 9.7: Rectified image pair (corresponding pixels are vertically aligned).

In Figure 9.8 three orthographic views of the reconstruction obtained from a single image pair are shown. These allow to verify the metric quality of the reconstruction (e.g. orthogonality and parallelism).

Figure 9.8: Orthographic views of a reconstruction obtained from a single image pair: front (left), top (middle) and side (right).

To have a more complete model of the reconstructed street it is necessary to combine results from more than one image pair. This could for example be done using the volumetric approach presented in Section 8.2). A simpler approach consists of loading different surfaces at the same time in the visualization software. There is no need for registration since this was automatically performed during the structure and motion recovery. Figure 9.9 contains four views of a model consisting of 7 independently reconstructed 3D surfaces.

Figure 9.9: Views of a reconstruction obtained by combining results from more images.