Research On Vision-based Environment Modeling And Autonomous Landing Of Unmanned Aerial Vehicle Systems

The position,attitude,speed and other state information of the UAV is generally obtained by GPS(Global Positioning System)and airborne sensors.However,GPS signals cannot be obtained in many scenes.The method of positioning estimation using airborne sensors only can greatly improve the autonomous ability and environmental adaptability of the UAV.In addition,although the monocular SLAM(Simultaneous Localization and Mapping)algorithm can obtain the pose estimation of the camera in real time,it cannot obtain the depth information of the scene,so it cannot directly provide state information for the UAV.Finally,in the process of landing,if there is no reliable autonomous landing method,the UAV may fail to land or even crash.Aiming at these problems,this paper studies the visual environment modeling and autonomous landing method of UAV system.The main work is as follows:(1)Autonomous construction of UAV visual platform based on ROS(Robot Operating System).The first is the construction of hardware platform,including the construction process and communication mode among various parts such as UAV platform,airborne computer,vision camera and image transmission system,which provides a good hardware foundation for the implementation of the control scheme.Then,the software platform is built,through the design of pose estimation,motion planning,trajectory tracking and autonomous landing algorithms to achieve the precise control of the UAV;Finally,the successful construction of the UAV vision platform is verified by running UAV demonstration routines on the simulation platform.(2)Design of UAV monocular SLAM extensible framework with deep recovery capability.In this method,a scalable SLAM framework with depth recovery function is designed for UAV using only monocular camera sensors.Firstly,the UAV pose information with proportional scale is obtained based on ORB?SLAM2 algorithm,and the Apriltag2 algorithm is used to perform fusion operation to recover the scene depth information.In addition,based on the monocular sensor,the pose conversion module and pose publishing module are designed to provide the state information that can be directly used for the UAV.Finally,the outer ring geometry tracking controller of UAV is designed,and the flight task generated by trajectory planning module is completed by calculating the desired thrust and attitude Angle.The effectiveness of the proposed framework is verified by simulation and actual experiments.(3)Design of high precision UAV autonomous landing system.A high precision autonomous landing system is designed for UAV based on monocular camera sensor.Firstly,during the UAV flight,the VINS-Mono algorithm is used to provide real-time state estimation information.After detecting the landing tag,the UAV pose estimation information is optimized by the April Tag2 algorithm.Then the motion planning algorithm is used to generate the landing trajectory,and the optimized motion controller is used to drive the UAV to land accurately to the fixed target platform.In addition,in order to further expand the application scenarios of this scheme,a landing algorithm on the moving target platform is designed.Finally,the feasibility of the proposed method is verified by experiments of fixed landing target and moving landing target.