Evaluation of relative pose estimation methods for multi-camera setups

The fully automatic and reliable calculation of the relative camera pose from image correspondences is one of the challenging tasks in photogrammetry and computer vision. The problem has not been solved satisfactorily, e.g. in case of critical camera motions or when observing special point configurations. Furthermore, some methods provide multiple solutions for the relative orientation. In this paper we compare various techniques, analyze their difficulties and give results on synthetic and real data. We show that in case of noisy data pose estimation of a single camera remains difficult. The use of multiple calibrated cameras that are fixed on a rig leads to additional constraints, which significantly stabilize the pose estimation process. 1