Optimal Guidance Design for UAV Visual Target Tracking in an Urban Environment

This paper addresses a vision-based air-to-ground target tracking problem of an unmanned aerial vehicle. A navigation filter is designed based on an extended Kalman filter to simultaneosly localize a moving target and an own-ship UAV by fusing vision-based measurements and other onboard sensors. Particularly, it is suggested to utilize sparse optical flow estimation to aid the inertial navigation when GPS signals are inaccessible. A guidance law aims at making a UAV pursue a target's horizontal position while vertically avoiding obstacles whose global positions are known. In order to achieve this guidance purpose, high accuracy in both the self and the target localizations is required. Since the vision-based navigation performance significantly depends on the camera motion relative to objects of interest, this paper proposes a stochastically optimized guidance law which attains target tracking and obstacle avoidance while maximizing the navigation accuracy. For real-time applicability, the one-step-ahead optimization technique is applied to derive a suboptimal guidance solution. Simulation and flight test results are presented to demonstrate the advantage of using the suggested optimal guidance policy instead of a nominal linear guidance.