Statistical estimation for exterior orientation from line-to-line correspondences

Abstract

This paper presents a statistical estimation from which a new objective function for exterior orientation from line correspondences is derived. The objective function is based on the assumption that the underlying noise model for the line correspondences is the Fisher distribution. The assumption is appropriate for 3D orientation, is different from the underlying noise models for k pixels positions, and allows us to do a consistent estimation of the unknown parameters. The objective function gives two important facts: its formulation and concept is different for that of previous work, and it automatically estimates six unknown parameters simultaneously. As a result, it provides an optimal solution and better accuracy. We design an experimental protocol to evaluate the performance of the new algorithm. The results of each experiment shows that the new algorithm produces answers whose errors are 10%–20% less than the competing decoupled least squares algorithm.