Computation of the Quadrifocal Tensor

This paper gives a practical and accurate algorithm for the computation of the quadrifocal tensor and extraction of camera matrices from it. Previous methods for using the quadrifocal tensor in projective scene reconstruction have not emphasized accuracy of the algorithm in conditions of noise. Methods given in this paper minimize algebraic error either through a non-iterative linear algorithm, or two alternative iterative algorithms. It is shown by experiments with synthetic data that the iterative methods, though minimizing algebraic, rather than more correctly geometric error measured in the image, give almost optimal results.