Flight Control Of Quadrotor For Panoramic View Construction

The quadrotor has light fuselage structure and flexible maneuverability and can take off vertically and hover in the air, which makes it very suitable to execute monitoring and detect-ing tasks near the ground. Because of limitation of payload, usually quadrotor can only get environment information from monocular camera. With image stitching method panoramic view can be constructed. To make sure that the panoramic view is integrated and consistent, quadrotor need to navigation as set route and get image data at set point.To make quadrotor have ability of autonomous flight control and environment awareness, the thesis builds quadrotor experiment platform with supporting software system. Upper computer on-board can collect and process data from sonar, optical flow, IMU and vision sensor. Optical flow module with sonar provides quadrotor with real-time velocity and alti-tude data, at the same time anti-shake camera can get sharp image when quadrotor is moving. Wi-Fi module make sure data of sensors can transmit to ground station. Upper computer and ground station software system can not only transmit data between quadrotor and upper com-puter and ground station, but also can control quadrotor with ground station autonomously.Hovering is basis of autonomous flight control. Without GPS or position marker, quad-rotor cannot get position relative to ground. With sonar and optical flow sensors altitude and velocity can be directly measured. So hovering of quadrotor is achieved by altitude controller and velocity controller which are based on PID. To make quadrotor flying as set route, posi-tion relative to ground need to be calculated. Quadrotor’s displacement can be integrated by velocity, and accumulation of displacement gets quadrotor’s displacement relative to ground. In this thesis, we design a cascade PID controller, in which the position output is given to ve-locity controller as desired value, and with velocity control quadrotor gets to set position.In order to construct panoramic view, we need to stitch the images. Because of rotated and scale invariance and low time cost of ORB (Oriented FAST and Rotated BRIEF) feature, we use image registration based on ORB feature to find out transformation matrix between images. In order to eliminate cumulative error, we use bundle adjustment to optimize trans-formation matrix. Finally multi-band blending is used to eliminate seams and ghosting in overlap area. Experiment results show that the proposed method can construct panoramic view without geometric dislocation or stitch mark when there rotation, zoom or brightness changes between images.