Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
Comparison of Filtering Methods for Stereo Visual-Inertial Navigation of Multirotor Aerial Vehicles
Submission Author:
Beatriz A. Asfora , SP , Brazil
Co-Authors:
Davi Antônio dos Santos, Beatriz A. Asfora
Presenter: Beatriz A. Asfora
doi://10.26678/ABCM.COBEM2017.COB17-1006
Abstract
Vision-aided inertial systems are thought to be a good fit for the micro multirotor aerial vehicle navigation issue, not only due to the camera cost, lightweight and low power consumption, but also for its ability to provide rich information about the environment. The present paper focus on the evaluation of three non-linear Kalman-based filters applied to the estimation problem of the position, linear velocity and attitude of an aerial vehicle, along with the biases of its inertial sensors. A non-linear dynamic model is proposed, allowing sensor fusion to be performed in a prediction-update framework, using measurements from a strapdown inertial measurement unit and a stereo visual system in a GPS-denied environment with four known landmarks. The extended Kalman filter, unscented Kalman filter and ensemble Kalman filter are compared regarding accuracy, computational burden and robustness with respect to initial conditions, in a simulated scenario using the same tuning parameters. Results from Monte Carlo simulations show that regarding accuracy and precision, the unscented Kalman filter proved to be equivalent or superior to the other two in all configurations that were tested. As expected, the extended Kalman filter remains as the computationally lighter approach, but not the most robust one: in this matter, the still not fully explored ensemble Kalman filter excelled.
Keywords
Visual-inertial navigation, multirotor-aerial vehicles, Sensor Fusion, Kalman Filter
