Pose estimation using geometric constraints

Abstract

The paper concerns 2D-3D pose estimation in the algebraic language of kinematics. The pose estimation problem is modeled on the base of several geometric constraint equations. In that way the projective geometric aspect of the topic is implicitly represented and thus, pose estimation is a pure kinematic problem. The authors propose the use of motor algebra to model screw displacements of lines or the use of rotor algebra to model the motion of points. Instead of using matrix based LMS optimization, the development of special extended Kalman filters is proposed. In this paper extended Kalman filters for estimating rotation and translation of several constraints in terms of rotors and motors will be presented. The experiments aim to compare the use of different constraints and different methods of optimal estimating the pose parameters.