Vision-Based Runway Recognition And Pose Estimation For UAV Autonomous Landing

In order to improve the autonomy and intelligence of UAV landing, fixed wing UAV landing method based on vision is researched in this paper. This research is consist of three parts:runway recogniton, runway tracking and pose estimation,which is with the help of image squences photographed by an UAV when landing.The runway recognition method is presented based on Sscale-Invariant Feature Transform (SIFT) matching. The SIFT feature database of runways is build firstly, then to match the SIFT feature extracted in the real-time image photographed by the camera of UAV, when the it is to land on the runway in some area. If the matching succeed, we known that the runway exists in the real-time image, and the runway position,orentation and scale in the image is known using corresponding relationship between points. Because of the consuming time of SIFT matching is large, after the runway is recognitioned, it should been tracked in order to achieve real-time ability. The runway tracking method is based on Fast NCC template matching mehtod. If the runway is positioned, we do edge detection and Hough thransform to extract edge line of runway, and then use extracted line in the image to establish line equation, and use projection geometry technology and pinhole principle to calculate the pose of UAV.The experiments show that using SIFT feature matching method, we can match hundreds of frame images successfully, and sovle runway recognition problem under complex environment; the runway tracking mathod achieved accuracy of 100%, which is robust to environment or illumination change; the pose estimation result is conform to reality of UAV landing. So the reseach of this paper is useful for UAV landing based on vision.