Research On Autonomous Landing Technology By Visual Information For UAV

UAV currently widely used in various fields, especially in the military filed since it’s military value prompt the country from all over the world rushing to develop UAV’s technology. It is made up of take off, recognition, monitoring, tracking and landing. In recent Years, researchers around the world focus on different technology according to different features of UAV. The article simulates the technology of identification, tracking and landing for UAV.The experiment takes the H landmark as the research object. During the tracking stage in the experiment, we provide a scheme for locating target by incorporating the probability classifier and SURF points’tracking through the landmark’s SURF and geometric features. We estimate the attitude of UAV by matched SURF points. The strong classifier is composed of several weak probability classifier based of integral image. SURF tracking is on the basis of SURF points and contour characteristics. We can accurately position the target by the fusion rule of the probability classifier and SURF tracking. When there is only one success find the position of target, we can judge weather the candidate is searched target through matching it with the detected object stored in memory. The system can be divided into early selection, classifier training and target tracking. The stage of early selection get a attitude which is relatively parallel with target and provide a competitive target feature for the following stage. Experiments shows that the system is with a good robustness (rotation angle, pitching angle and light) and can meet the requirement of real-time processing. At the same time, we can get the attitude of UAV through matched SURF points bounding in target.Although the system with landmark can make a good performance in landing by recognition and tracking, but the UAV must be with the capability dealing with the emergency circumstances. For example, the situation is that landing without landmark. Using binocular vision system can solve it being without scene environmental information. The paper builds the binocular system through calibration, polar rectification, image match, ENCC disparity, scene reconstruction and selection of landing area. During the stage of scene reconstruction, we optimize the matching state by SURF point tracking. The article introduces contour matching, SURF tracking, ENCC stereo matching and RANSAC curved surface fitting for the system. ENCC sub pixel stereo matching can obtain higher accuracy of parallax figure and provide better data to select safe area for UAV. Contour matching can determine the safe area quickly and effectively and select enough SURF points to fit the plane through RANSAC. SURF tracking reduces running time and effectively tracks target. Experiments show that the proposed algorithm can efficient real-time provide safe landing area for UAV.