Linear Projective Reconstruction from Matching Tensors

This paper describes initial work on a family of projective reconstruction techniques that extract projection matrices directly and linearly from matching tensors estimated from image data. The simplest methods use fundamental matrices and epipoles, alternative ones use trilinear tensors. All of the data is treated uniformly, and there is no reliance on ‘privileged’ images or tokens. The approach is based on ‘joint image closure relations’ — bilinear constraints between the projection matrices and their matching tensors, that express the fact that the latter derive from the former. The performance of the new techniques is quantified and compared with that of several existing methods.