Comparative Evaluation of Filtering Methods in an Experimental Visual-Inertial Attitude Determination Setup
B. A. Asfora1; E. A. Moura1; D. A. Santos1
1 Instituto Tecnológico de Aeronáutica
doi:10.20906/CPS/CILAMCE2017-0695
Resumo
The small size and fast dynamics of Micro-Multirotor Aerial Vehicles impose challenges regarding their navigation and control. Visual-Inertial schemes are a good fit for tackling the navigation issue, and although Monte Carlo Simulations provide valuable information on sensor fusion methods, it is fundamental to have a controlled real-world comparative basis for practical implementation of the algorithms. This work is concerned with the experimental comparison of the Extended Kalman Filter, Unscented Kalman Filter and Ensemble Kalman Filter. In the setup presented, they each rely on the measurements of a three-axis rate-gyro and an upward-facing camera to estimate the platform attitude at each instant based on the known positions of landmarks. The platform is a Quanser 3-DOF Hover kit, with high accuracy encoders that provide an alternative measure of attitude, taken as the truth state. The performance of each filter is evaluated in a similar manner as it would be in computational simulations: they can be tested in a standardized but real environment, providing a complementary analysis, so there can be a mindful choice of the more suitable approach for each goal.
Palavras-chave: MAV; attitude determination; visual-inertial navigation; experimental analysis; Kalman filter