Vision-based ground target tracking using guidance filter for a rotary UAV

Abstract

This thesis presents a target tracking approach for a miniature rotary unmanned aerial vehicle(UAV) using only and onboard gimbaled(pan-tilt) camera(vision) system. In this thesis, vision-based target tracking problem is divided into four phases: the target pointing algorithm, the estimation of the target``s position and velocity, the tracking guidance law, and the guidance filter. Target pointing is performed as centering the target``s feature on the image plane. For rotating a camera, gimbal system which has pan-tilt rotation was adopted, and the algorithm was verified through the actual flight tests. Out of a range of the gimbal``s angle, to maintain the target within the camera field of view, a flight guidance law is necessary. In this thesis, as a tracking guidance law, the chasing guidance law was proposed. The guidance law is performed that UAV``s heading tracks the angle that points the target. For performing the chasing guidance law, the estimation of the target``s position and velocity is indeed required. For solving this problem, Extended Kalman filter(EKF) and Unscented Kalman filter(UKF) were adopted. The process model of the filter is derived through the UAV velocity, the target velocity, the pin-hole camera model(pixel coordinates), and the image depth (relative altitude) obtained by transformation coordinates. The multiple guidance filters combined with a guidance law and the filters enable to perform the chasing guidance under the target``s various motions such as constant velocity, acceleration, deceleration, stop, and curving. Finally the performance of the guidance filter was verified by using simulations.