Autonomous Landing Of An Unmanned Aerial Vehicle Based On Computer Vision

Unmanned helicopter, due to its good performance, easy maintenance, low cost and many other advantages, has been widely used in military and civilian areas. Recently, many institutions both at home and abroad take it as an important research object. Independent landing is an important component of helicopter autonomous flight. Independent landing based on computer vision is one of the main research contents in the field.According to the geometrical characteristics of traditional H landmark, this paper presents a method of landmark recognition based on contour fitting, which judges the landmark's situation by computing the numbers of landmark contour's vertex. If the image contains full landmark, real-time calculation of corners can obtain the helicopter's attitude information. If the image contains part of the landmark, the helicopter's attitude information can be derived from real-time calculation of corners. And if the image contains only part of the landmark, the moving direction and displacement of the helicopter, which would make the landmark presented in the view completely, can be estimated. Experimental results indicate that the method has good real-time performance and stability, and the purpose of presenting full landmark could be realized through the displacement'estimate effectively.Considering the independent landing under the case without landmark's guidance, monocular vision can not complete the attitude's calculation of unmanned helicopter for lack of priori knowledge about the environment. Therefore binocular vision system is built to reconstruct the current scene. The regional flatness and tilt angle with respect to the camera plane are calculated in order to judge whether the current region meets the projected landing area. After we get the region satisfying conditions, the region could be tracked through matching points between neighboring frames. With the matching points, the region's angle and position relative to the camera plane could be real-time computed.