Vision-assisted Auto Landing Of UAV With Onboard Embedded System

Small multirotor UAV(unmanned aerial vehicle) is more and more popular these years. It is widely used in aerial photography, surveying and mapping, inspection of electric infrastructure and aerial surveillance. The growing applications of UAV have produced an urgent need for the intelligence of UAV. The computer vision which possesses diverse information and independence provides the UAV with a large amount of information about the outside world. Therefore, the research to the vision navigat io n of UAV has a profound meaning to the development of the UAV intel igence.Firstly, the basic movement status of a quadrotor is il ustrated. On the purpose of extending the flight time, multiple approach is discussed and an efficiency experiment for the motor-rotor combination is conducted. Based on the efficiency data of motor-rotor system, multiple quadrotors with different takeoff weight, varying time of flight and reasonable payload are built. The mathematical model of a quadrotor is developed using Newton-Euler equation. The equation of state is also developed. The model for the motor and rotor is established using a first-order inertial elements.Then, the UAV platform for the vision navigation is designed and established. Guided by the principle of universality, load balancing and reliability, a new architecture of flight-control system with two processors is proposed. The subsystem of flight-cont ro l system is also designed and the onboard sensor selection is finished. An onboard image processor is selected based on the computing need of the image algorithm.Afterwards, the control system and image processing method is mainly discussed. To get the vision feedback, the camera model and expression of the camera attitude/ position is developed. Then, the relative estimation method for attitude/position is developed. A two-stage auto-land control method is proposed. A gimbal track system is also designed to help the camera track the landing target. Three auto-land procedures are introduced based on the difference in the hardware structure and control method. At last, some improvements to the image processing method and control method is introduced to solve the last-stage loss of vision feedback.Finally, experiments, including aircraft attitude control, speed control and position control is conducted. The attitude controller, speed controller and position controller show excel ent performance. The high-precision auto-land experiment is also successful. The whole system is working normally and verifies the system design.