Photogrammetric relative pose estimation with varying and unknown focal length

This paper addresses on the estimation problem of relative pose from two views with the varying and unknown focal length. For statistical optimization, approximated maximum likelihood estimation in the presence of noise and measurement model is proposed under the inter-image epipolar constraint. In order to achieve a direct solution of relative pose, the fundamental matrix is reparameterized with minimal degrees of freedom. Then, we derive the analytical differentiation and adopt the Levenberg-Marquardt algorithm to solve the optimization problem for efficient computation. Experiments show that the proposed method is comparable to bundle adjustment while providing a significant advantage in terms of computational cost.