I wanted to recreate the results of this paper, where the authors draw the boundary between two parts of a building from stereo images. The authors mention that the epipole and the intersection of the two planes (two sides of a building) from two cameras can be found from the eigenvectors and the eigenvalues of the homology, and use this to plot some nice images in Figure 4 whereby the building can be divided by the planes.
When I try to reproduce this though, I am unable to recreate the results. My intersection is completely wrong. I tried to recreate by first finding the eigenvectors of the pair of similar eigenvalues, which according to the paper correspond to the intersection line of the planes. Then I take the cross product of the eigenvectors to get the axis, and then draw this line to my image by getting its intersection with the borders using the dot product. My line however seems to bear no relation at all to the real intersection. Is there another step that isn't mentioned here that I am missing?