Self-calibration from four views with possibly varying intrinsic parameters

In this paper we address the problem of evaluating how accurate the Euclidean reconstruction using a self-calibration method can be. We focus on the self-calibration of a binocular head-eye system from four views with varying intrinsic and extrinsic parameters. The only information provided to the algorithm is the fundamental matrices [1] and the point correspondences between the four views. The main improvements of this article are: (a) to take account of intrinsic and extrinsic parameters change; (b) to exploit the information of the cross-correspondences to improve the Euclidean reconstruction.