| dc.description.abstract | This thesis presents design and comparison of a vision-aided uniformly semi-globally exponentially stable (USGES) nonlinear observer (NO) and a Multiplicative Extended Kalman Filter (MEKF) for estimation of attitude, gyro bias, position and velocity of a fixed-wing Unmanned Aerial Vehicle (UAV). The NO uses measurements from an Inertial Measurement Unit (IMU), a Global Navigation Satellite System (GNSS) receiver, and a video camera. The MEKF and the NO are evaluated with real world experimental data. Moreover two new NO representations are proposed. The proposed NOs have a computer vision (CV) system that is based on epipolar geometry, and hence independent of the depth in the images and the structure of the terrain being filmed. The first proposed NO utilizes a camera and the continuous epipolar constraint and is named Continuous Epipolar Optical Flow Nonlinear Observer (CEOF NO). The second proposed NO uses a camera and the epipolar constraint and is named Epipolar Optical Flow Nonlinear Observer (EOF NO). Experimental data from a UAV test flight show that the vision-aided NO is substantially less computational demanding than the MEKF. It is seen that the NO has similar performance as the MEKF by means of accuracy of the estimates. The NO and the MEKF estimates are compared with an Extended Kalman Filter (EKF) implemented on the onboard autopilot of the UAV. The results illustrate that the estimates of the states converge accurately to the correct values. Moreover simulated data show that the proposed observers have more robust CV than the previous developed vision-aided NO, yielding more accurate and robust performance. | |
