Research On The Key Technologies For UAV Landings Based On The Aid Of Computer Vision

Computer vision has become an important source of information in autonomous navigation of UAV with its characters of economy, passivity and information abundance. In this paper, we take image sequences of ship model captured by binocular camera to study.The method for UAV autonomous landing is researched which consist of the identification and tracking of UAV, the extraction of landing field feature and the relative pose estimation of UAV by the integrated use of digital image processing, computer vision and 3D reconstruction techniques.Firstly, we use the method combined SIFT feature matching and template matching to solve the problems of identification and tracking in conditions of complex background. The SIFT features library of ship runway is established, then we do the match between the SIFT features library and the images captured by the airborne cameras. A threshold is set so that we can determine whether the target has appeared, and the runway area will be located if the ship has been identified. In order to achieve real-time tracking, a template matching algorithm is used with the strategy of template refresh being made.Illumination can have a significant impact on the effective segmentation of the landing area image. An image enhancement algorithm is proposed, which combines image restoration techniques based on the average of values of images? grayscale and image enhancement technology with linear grayscale transformation, to eliminate the adverse effects of illumination changes. An iterative method of these two techniques which relies on the feedback information is made. Simulation results show that the algorithm performs well under variable conditions of illumination, and the characteristics of runway lines are highlighted.The binary image is achieved after the application of the methods on the enhanced images of runway area which are gray stretch and threshold segmentation. Then we use Canny operator to detect the edges, and the lines in the images are founded by the method of probabilistic Hough transformation. In order to select the right lines to compute which are the outer edges of two runway lines, a screening stragdy is made. On this basis, the four points on the outer edge of the runways are reconstructed based knowledge of projective geometry and three-dimensional reconstruction, and then the flight parameters of UAV include position and attitude are worked out with Kalman filtering which can fix the wrong computation.The results show that the purpose of identification and tracking of the ship is achieved in varied background and perspectives, which indicate the robustness of the method. And then, the outer edges of the runways are picked up precisely from the located area of the image. Besides, a sliding platform is established for t he simulation of the process of UAV landing. The results show the effectiveness and feasibility of the algorithm by compraring the experimental data and the real data measured by IMU.