A Study Of Integrated Navigation Technology Of MEMS Inertial/Monocular Visual Odometry

Whether it is unmanned vehicles,unmanned aerial vehicles or unmanned underwater vehicles,its accurate independent positioning is an important prerequisite for its realization of autonomous navigation.The traditional navigation methods are applied to land-space GPS satellite navigation,laser radar for indoor applications and underwater acoustic navigation.However,these navigation methods are not suitable for the land,sea and air fields.The combination of IMU and vision sensor is low cost,compact and complementary,is a more common solution.Their complementarities are mainly reflected in the following aspects:Integrating the noisy angular velocity and acceleration information measured by the Inertial Measurem Unit(IMU)twice makes the pose of the noise amplified after prolonged amplification.On the other hand,IMU estimation is relatively accurate in a short period of time,and motion blur occurs when the camera is moving,which is a weakness of the visual sensor.Compared with the IMU,camera data can not be substantially drifted,and camera data can be effectively estimated Correct IMU drift.In this thesis,a low-cost and high-precision autonomous navigation system,Visual Inertial Odometry(VIO),is proposed.This system implements a tightly coupled combination of inertial visual information using an unscathed Kalman Filter(UKF),The visual information can be used as the observation equation without using the depth information of the feature points to construct the geometric constraints of the three-dimensional space by directly using the two-dimensional feature pixels of three consecutive frames of images.The following is the main work of this article are as follows:(1)Compare the effect of three-sample compensation method and Runge-Kutta method on attitude solution.Inertial navigation solution of the pose accuracy is basically determined by the attitude algorithm.When there is noise interference in the system,the accuracy of the attitude solution of the two algorithms is compared by MATLAB simulation.Thus,the fourth-order Runge-Kutta method has better anti-interference performance under coning motion.(2)The constraint equations of the pole geometry and the triple focus tensor are briefly deduced and the state equations of the VIO system are established based on this.In computer vision geometry,the pole geometry constraint reflects the inherent geometric relationship between the two views.This thesis derives from the geometric perspective.The three-focalpoint tensor reflects the intrinsic relationship between the three views.In this theis,we derive from the perspective of algebra This inner geometry of the three views.(3)Aiming at the characteristics of VIO observations,a new RUKF nonlinear filtering algorithm based on Unscented Kalman Filter(UKF)based on Random Sampling Consistency(RANSAC)is proposed.The algorithm realizes the tight coupling between IMU and monocular vision,and achieves the matching of the matching feature points.The precise information of pose and pose is obtained by the fusion of refined and refined matching points and IMU data.This thesis first verifies the superiority of UKF filtering algorithm over EKF filtering by simulation and then verifies the VIO state space model presented in this thesis by using the RUKF algorithm proposed in this thesis in KITTI sports car data set.The results show that the proposed VIO system The feasibility and superior features compared to the binocular vision positioning system.