Robust metric velocity estimation for non-overlapping camera systems

This paper introduces a novel, robust and efficient approach for metric velocity estimation using multiple cameras with no overlapping fields of view. By incorporating rotation priors, the metric velocity estimation problem reduces to solving a set of linear equations. To determine the metric velocity, a total of three temporal point correspondences provided by at least two cameras are required. This enables the use of a 3-point RANSAC algorithm for robust outlier rejection. To enable the use of the algorithm on platforms with different computational budgets, we present three approaches with increasing computational requirements to solve for the metric velocity. The scheme can detect if the motion of the camera system is close to degeneracy and degrades gracefully. The proposed scheme is evaluated using synthetic image sequences.