A globally exponentially stable non-linear velocity observer for vision-aided UAV dead reckoning
Lorenzo Fusini, Tor Arne Johansen, Thor Inge Fossen · 2016
This paper introduces a non-linear observer for dead reckoning of a fixed-wing UAV. The sensors used are an Inertial Measurement Unit (IMU), a camera, an altimeter, a magnetometer, and an inclinometer. A machine vision system calculates the optical flow (OF), which is used to provide the body-fixed velocity of the UAV. The observer is able to estimate attitude, gyro bias, vertical position, linear velocity, and linear acceleration, whereas the horizontal position is calculated by direct integration of the estimated velocity. A key contribution of the paper is the proof of global exponential stability (GES) of the observer, together with a test on experimental data: the horizontal position is calculated through dead reckoning, which leads to an inevitable divergence from the real values, but the exponential convergence of the linear velocity guarantees that the position can diverge at no more than a bounded rate. Experimental data are used to test the observer and compared with the output of an extended Kalman filter (EKF): the estimates converge quickly to similar estimates as the EKF, and the horizontal position is shown to drift at no more than a bounded rate, as expected.