Projective factorization of planes and cameras in multiple views

This paper proposes a novel method for the projective reconstruction of planes and cameras from multiple images by factorizing a matrix containing all planar homographies between a reference view and all other views. If some planes are not visible in all views an alternative method is presented which solves the problem in two steps: a) all camera centers are recovered simultaneously b) all planes are reconstructed. The key idea of both methods is to specify one of the planes, which is visible in all views, as the plane at infinity. The methods were applied to synthetic and real data, where VRML models can be created with a small amount of user interaction.