Direct feature correspondence in vision-aided inertial navigation for unmanned aerial vehicles
F. Paredes-Valles, D. P. Magree, E. N. Johnson
International Conference on Unmanned Aircraft Systems (ICUAS), 2017
Abstract
This paper proposes a novel method for corresponding visual measurements to map points in a visual-inertial navigation system. The algorithm is based on the minimization of the photometric error on sparse locations of the image region, and realizes a gain in robustness that comes from the elimination of the need of feature-extraction for correspondence. The system is compared to a standard approach based on feature extraction, within a visual-inertial EKF formulation. High-fidelity simulation results show the proposed method improves the horizontal RMS error by means of increasing the number of features corresponded by the algorithm.