Research On Indoor Autonomous Flight Of UAV Based On Monocular Vision

With the development of UAV technology,the application scope of UAV is more and more extensive.The usage scenarios are also developing indoors.The key to indoor UAV flight is indoor positioning and navigation and obstacle avoidance.In indoor environments,the location information sources of UAVs are very limited,which greatly limits the use of UAVs.Simultaneous Localization And Mapping(SLAM)technology can obtain location information using only the sensors onboard the drone.In this thesis,according to the key technologies required for the indoor flight of UAV,the indoor positioning algorithm and indoor navigation and obstacle avoidance of UAV are studied,and the UAV flight platform is built for experimental verification.Firstly,the SLAM system based on monocular vision is studied,aiming at the problems of poor stability and easy failure of the system in rotation and rapid motion in monocular vision SLAM.The Inertial Measurement Unit(IMU)is selected to make up for the insufficiency of visual SLAM and realize a SLAM system that integrates vision and inertial navigation.The IMU can also accurately obtain the change of the pose when moving rapidly,which can effectively make up for the insufficiency of pure visual SLAM.The combination of the two can effectively improve the stability and accuracy of the system.In the fusion system,the IMU information between two frames is pre-integrated,and the result of pre-integration is used as the initial value of the state,and the camera and IMU data are combined to estimate the pose.In the local optimization,the sliding window method is used to optimize the window.’s pose.Verified by the dataset,the results from the V2＿03 dataset show that the stability of the SLAM system in this thesis is significantly better than that of pure visual SLAM in fast-moving scenes.And the absolute pose error of the SLAM system in this thesis is also smaller than that of the pure visual SALM system.After that,the indoor obstacle avoidance algorithm of UAV is researched,and the obstacle avoidance algorithm in the two-dimensional laser map is selected.The dynamic window method is a classic and widely used obstacle avoidance algorithm,which finally selects the appropriate forward direction and speed by sampling and scoring the possible speed of the robot.The dynamic window method is improved when the obstacles are dense,and the landmark points are selected from the global path as guidance,so that the robot can better bypass the obstacles.Finally,a set of UAV indoor flight platform was built,which cooperated with visual inertial navigation SLAM and laser map to complete the indoor flight task of UAV.Using visual inertial navigation SLAM to obtain the position information of the UAV,and using the laser map for path planning,the UAV has the basic ability of indoor flight,and it is verified in the simulation environment and indoor scene.