A New Line-Based Orthogonal Iteration Pose Estimation Algorithm

Abstract

The Orthogonal Iteration (OI) algorithm for pose estimation from point correspondences is global convergent, efficient and robust, which make this algorithm quite attractive. In this paper, we extend the OI algorithm to 2D-3D line correspondences and present an iterative algorithm referred to as LBOI (Line-Based Orthogonal Iterative Algorithm). We first derive a new form of fundamental equations and get an error function which has actual geometry meaning. We then divide the error function into two parts and minimize both of them to estimate the best pose parameters iteratively. The iteration process of our method is similar as that of the OI algorithm. However, the rotation matrix is computed twice in each iteration in our method. For different groups of line correspondences, we present different methods for getting an initial guess. Synthetic experiment data show the proposed algorithm is insensitive to initialization, efficient and accurate.