Four-rotor UAV Navigation Research Based On Stereo Visual-intertial SLAM

The simultaneous localization and mapping(SLAM)takes the camera as the main sensor to estimate the camera pose and build environment map,which makes great progress in the autonomous flight of UAV.In this paper,aiming at the problem that the tracking of slam system based on visual features is easy to fail when the image is fuzzy,the motion is too fast and the features are insufficient,a stereo vision inertial SLAM(VI-SLAM)system based on tight coupling and nonlinear optimization is proposed.Firstly,the position and attitude of key frame are used as constraints to initialize the deviation of inertial measurement unit(IMU)and predict the position and attitude of current frame according to the IMU pre-integration.Then,in the back-end optimization,the nonlinear local smoothing method is used to fuse the position and pose estimation of visual slam with the pre-integration of IMU,and the accumulated error is eliminated by closed-loop detection to keep the global map consistent.Usually,the sparse map constructed by visual SLAM algorithm can not be used in autonomous flight of UAV,this paper extends the sparse feature point map to the dense octree map.The performance of the system is validated by EuRoC dataset.In this paper,the nonlinear optimization VI-SLAM algorithm based on tight coupling is about twice as high as that of the visual SLAM navigation algorithm.The algorithm is applied to the Four-rotor UAV platform,and the validity and robustness of the algorithm are verified.The main work and achievements of this paper are as follows:1.The image acquired by the camera contains rich environmental information,but it is sensitive to fast motion;the frequency of IMU sensor is high and its dependence is small,but there is measurement drift;the complementarity of the two is suitable for fusion.Based on the ORB-SLAM2,a stereo vision inertial navigation slam system is proposed in this paper.Through multithreading parallel processing,the map can be closed-loop and reused.2.In order to solve the bias and noise of IMU in the stereo vision inertial SLAM.Firstly,the pose of the key frame is used as the constraint to estimate the bias of the IMU.Then the pose of the current frame is predicted according to the pre-integration of the IMU as the initial value of the estimated pose.Because the depth information of feature points can be calculated by binocular camera,the computation complexity of IMU initialization is reduced.3.In this paper,a tightly coupled stereo vision inertial navigation SLAM system with nonlinear optimization is proposed.In the process of optimization,the pose estimation of visual SLAM and the constraint of IMU pre-integration are considered simultaneously.The nonlinear local smoothing method is used.In the sliding window,the pose of key frame and the pre-integration of IMU are optimized,and the deviation of IMU is updated.Finally,the accumulated error of the system is eliminated by closed-loop detection,and a global consistent map is obtained.4.In this paper,the method of map construction based on binocular camera is improved,which is extended from sparse feature map to dense octree map.It solves the problem that the sparse feature map created by ORB-SLAM2 system can not be used for UAV navigation.Finally,the euroc data set is used for the experimental evaluation.The performance of the algorithm in practical application is verified.