Sensor motion refinement through viewpoint correction and robust interpretation of normal flow

An algorithm is presented for reducing the uncertainty in existing estimates of a rapidly translating camera´s velocity. The algorithm combines the uncertain velocity information and low cost image motion estimates in order to produce a more accurate estimate of camera velocity. A model for the principal failure mode of the algorithm and the derivation of an acceptance test for the algorithm output are described. The performance expected from use of the acceptance test is compared with results from experiments with image sequences taken from an airborne camera