Vision-Based Online Localization and Trajectory Smoothing for Fixed-Wing UAV Tracking a Moving Target

Yong Zhou, Dengqing Tang, Han Zhou, Xiaojia Xiang, Tianjiang Hu

Abstract: Using unmanned aerial vehicle (UAV) to estimate and predict the position and motion of the ground target has been widely focused on in many computer vision tasks. This work aims to address the development of a vision-based ground target localization and estimation for a fixed-wing UAV. Limited by the lightweight onboard processor, it is conflicting with the need for online onboard operation and the computing resource limitation of the platform. In this paper, we develop a practical approach to recover dynamic targets based on extended Kalman filter (EKF) for localization and locally weighted regression for trajectory smoothing. Our methods run online in real time with the only data up to the current timestep. The flight experiment results show the effective tracking and localization of the ground moving target.