Research On Monocular Visual Inertial SLAM System Based On Point And Line Feature

Visual Inertial Simultaneous Localization and Mapping(VINS)is a key technology for achieving real-time localization of mobile devices in unknown environments,with applications ranging from mobile robotics,UAVs,augmented reality and autonomous driving.At present,point feature-based VINS techniques have become relatively mature,but they are usually insufficient to extract enough point features to ensure continuous tracking and highprecision localization in environments such as weak textures.There are abundant line features in real scenes,and they have geometric structure constraints that can be used to compensate for the lack of feature point constraints.In order to improve the localization accuracy and robustness of VINS systems in challenging environments(weak textures,fast motion,and dimness,etc.),a monocular visual inertial SLAM system based on point-line features is proposed in this paper.The main research methods and results of this paper are as follows:1.An adaptive gamma correction algorithm is used to dynamically adjust the brightness of the input image before feature tracking.In the line feature tracking thread,line feature extraction is performed using the improved EDLines algorithm,and line feature matching is performed using the proposed line feature matching algorithm based on the pyramidal LK optical flow method.2.The line features in the image are used to reconstruct the spatial landmarks and define the reprojection error of the line features.The Pn P algorithm is used to minimize the reprojection error of point features and line features to solve the inter-frame pose transformation and complete the pure visual initialization.In the tightly coupled nonlinear optimization model,the state variables are optimized by minimizing the objective function containing the IMU residuals,the reprojection error of the point feature and line feature.The five modules of data preprocessing,initialization,local VIO,re-localization and loop detection,and pose graph optimization constitute a complete monocular visual inertial SLAM system based on point and line features.3.An evaluation of the VINS system researched in this paper was conducted using the public EuRoC MAV dataset and the public TUM-VI dataset.By comparing with other advanced VIO and VINS systems,it is verified experimentally that the VINS system researched in this paper has higher localization accuracy,robustness and real-time performance.