A runway tracking model using Zernike moments and particle filters for a landing unmanned aerial vehicle based on vision

Abstract

An advanced runway tracking model for landing of Unmanned Aerial Vehicle (UAV) based on vision is proposed. This model builds on existing work, but extends it to achieve efficiency, robustness, and address some critical situations such as instant sun glare, instant heave fog, cloud hold back, and instant extinction of approaching marking, and so on. These situations always have bad effects to our visual landing system of UAV. So, two different schemes containing several approaches constitute the core of our visual system to address these situations. We use Zernike moments as a regionbased shape descriptor of runway and save the changing pattern through landing process of pretest. At the real flight time, we use particle filter to track the change of the Zernike moments that calculated on each potential region of runway at each frame. When this change is too big, exceed the threshold, we use the pretest data to reconstruct the shape of the runway. The performance of the presented schemes has been assessed throuth processing several video sequences that captured by the real landing plane. The experiment shows, this tracking model is more efficient and robust and can be used on a vision sensor for landing equipment of UAV or for an aerial vehicle's aided system.