Robust vision-based pose estimation for relative navigation of unmanned aerial vehicles

In this paper, we improve the accuracy and robustness of nonlinear least squares algorithm in pose estimation problem for UAV. To improve accuracy and robustness, first we reduced the noise of feature position of beacon. We apply Kalman Filter to feature position. After the Kalman Filter, the accuracy is improved approximately 40% in simulation study. Second, We organized the Relative Navigation Filter. To compose relative navigation filter, relative attitude kinematics and relative position equation are adopted. Using this filter, we could estimate relative velocity additionally and the accuracy was improved. And then, to improve the robustness we need appropriate initial state. The initial state estimation is based on linearization.