Multiple nonoverlapping camera pose estimation

Abstract

In this paper, we solve the pose estimation problem in real time using multiple nonoverlapping cameras and the Extended Kalman Filter (EKF). Four cameras arranged in two back-to-back pairs are put on the platform of a moving robot. The two axes passing through the camera centers of each pair are perpendicular. This arrangement aims to maximize the benefits of the back-to-back setting whose accuracy is shown in literature. Each camera has its individual EKF for pose estimation which enables accurate short base-line feature tracking and parallel processing. A model for multiple nonoverlapping cameras is formulated which improves the estimate of rotation parameters with the help of a median arbiter. Accordingly, the translational parameters of pose are estimated accurately and the scale factor ambiguity related to single camera methods is solved using a low-dimensional speedy optimization.