A Research On VIO-System Method Based On Point Line Feature Extraction

With the vigorous development of high-tech electronics industry and the improvement of computer technology,robots have been rapidly developed and applied in various fields,and visual SLAM(Simultaneous Localization and Mapping)can provide robots with their own motion information and The map information of the unknown environment makes it an important part of the navigation and positioning system in the field of autonomous robots.Visual SLAM also has the disadvantages of low data update frequency,not suitable for high-speed moving carriers,and too dependent on environmental conditions;while MEMS IMU can stably output high-speed moving carrier motion information,but there are cumulative errors,and the two have natural complementarity.However,visual SLAM based on point features cannot output accurate motion information in scenes with insufficient texture features,and the accuracy of visual navigation technology decreases when the body moves at a fast speed.Based on this point,the research object of this paper is the VIO system based on point and line feature extraction.The main research contents are as follows:Firstly,the imaging model of the monocular camera,the calibration method of related parameters,and the measurement model and kinematics model of the IMU are studied,and the camera calibration experiment is carried out.Design monocular vision front-end system through feature points and lines.Specifically,the FAST corner point feature information is extracted,the LSD algorithm is used to extract the line feature information,and the hidden parameters are modified and improved for the LSD algorithm to improve the calculation speed.The multi-layer pyramid optical flow method is used to match point features,and LBD descriptors are used to match line features to track the motion of features,and then the camera motion is recovered through the epipolar geometric relationship of feature points.Aiming at the insufficient robustness of pure visual SLAM,a discrete form of IMU data pre-integration based on median method is used to jointly initialize the system.The system is jointly initialized by the rotation and translation constraints between the IMU and the camera,and the velocity,position,attitude and scale information of the initial state of the system are obtained.In the back-end tightly coupled nonlinear optimization,the reprojection error of the visual line feature is added to the overall residual function of the Vins-mono system to optimize the state quantity of the system.In order to ensure the real-time performance of the system,the marginalization strategy of the sliding window is adopted to discard the old state variables,and the size of the sliding window is limited,so that the system can be solved at a high speed for a long time.Finally,the implementation of the entire VIO system is completed,and the public EuRoc data set is used for actual data verification,and the EVO tool is used as the evaluation of the trajectory solution error.