Research On Vision-based Navigation Key Techniques For UAV Autonomous Landing

UAV (Unmanned Aerial Vehicle) is the achievement of multidisciplinary development, and its navigation technology is dual-use technology, which has aroused national attention. UAV normally consists of Radar, or vision sensors, or inertial navigation, or GPS navigation system to navigate itself. The vision sensor has more advantages as it has the characteristic of light, low power consumption, and more over, it does not need inertial navigation or GPS, work in the passive mode, and easy to conceal. Research and development of visual navigation system is a key measure to improve UAV performance. The present research project is supported by the National Nature Science Foundation of China: Study of vision-based navigation key technique for UAV Autonomous Landing. The paper mainly focuses on UAV attitude parameter estimation and identification of runway.Firstly, binocular stereo vision method is adapted to estimate attitude parameters of UAV through loading. Based on UAV simulation platform, pitch angle and height of UAV are emphasized to study. After calibration of the images, the pitching angle is calculated based on the geometric relationships of vanishing line and focus of expansion. Sparse disparity map is got by extracting the corners, and then the height information is acquired, based on 3D reconstruction method, using pitch angle information. Further pitch angle is amended according to the statistical characteristics of the height.Considered real-time request of the vision-based on guidance system when the unmanned air vehicle is landing, we has studied support vector machine system for runway recognition based on small sample training. Through the gathered of runway image, the runway characteristic database has been established based on the statistical method. In view of different kernel functions, compared with has used the chromatic information the system performance, further studied and develops Bayesian classifier to supplement this system.In laboratory simulation environment, the UAV parameter estimation algorithm can effectively estimate the height and pitch angle parameter. In the VC++6.0 platform, runway identification simulation has been done and the experimental result shows that the algorithm can accurately identify the runway with fairly good robust, and can meet the real-time requirements.