Recursive Bayesian calibration of depth sensors with non-overlapping views

In this paper, we present a recursive Bayesian method to calibrate rigidly linked depth sensors with non-overlapping fields of view. The extrinsic parameters of this setup are obtained by rotating and translating both cameras, estimating the local transformations using point feature correspondences, and finally using these values to recursively find a solution to the matrix equation A_kX = XB_k. The algorithm is based on a Bayesian estimator, which allows the consideration of camera-specific measurement noise and permits the system to adapt naturally to changes in the extrinsic parameters. Special care was taken to keep the system free from singularities. This paper also includes a thorough evaluation based on synthetic and real data to show the effectiveness of the algorithm.