Least-squares estimation of anisotropic similarity transformations from corresponding 2D point sets

Pose estimation is a problem that occurs in many applications. In machine vision, the pose is often a 2D affine pose. In several applications, a restricted class of 2D affine poses with five degrees of freedom consisting of an anisotropic scaling, a rotation, and a translation must be determined from corresponding 2D points. A closed-form least-squares solution for this problem is described. The algorithm can be extended easily to robustly deal with outliers.

► We describe an algorithm for least-squares estimation of 2D anisotropic similarity transformations.
► The algorithm provides a closed-form solution that can be computed efficiently.
► The estimated transformation parameters are unbiased.