Point-Line Combination Feature Monocular Visual Inertial SLAM

Simultaneous Localization and Mapping(SLAM)is a method for robots to realize pose estimation and map construction in an unknown environment.It is an important branch of robotics and widely used in robotics,autonomous driving,virtual reality,etc.field.Visual SLAM(VSLAM)refers to methods for localization and map construction using images from visual sensors such as cameras.Compared with other SLAM technologies,VSLAM has the advantages of long detection distance,flexible path planning,and high positioning accuracy,so it has become a hot spot in current SLAM research.The traditional VSLAM mainly adopts the visual front end of point features.This method relies on the extraction of pixels with obvious pixel features and stable feature matching.However,this method has high requirements on the environment,and is easily affected by environmental factors such as illumination changes and feature sparseness,thus affecting the extraction and matching of feature points.Compared with point features,line features are more stable and less susceptible to environmental changes such as lighting,occlusion,and motion blur.Moreover,line features have rich geometric information,which can well make up for the missing geometric constraints between point features,thereby improving positioning accuracy and scene expression capabilities.Therefore,building a robust visual SLAM system should make full use of point and line features.This thesis proposes a monocular inertial SLAM method based on the geometric relationship between constructed points and lines.The work and research results of this thesis mainly include:1.A fast line matching method is proposed,which can improve the accuracy of line segment matching under image distortion changes by using the affine invariant relationship between points and lines,multiple false matching methods and global consistency scoring,and the time Less overhead.2.A stable map line tracking method is proposed,which uses six-degree-offreedom Pluck’s line coordinates at the front end,and four-dimensional orthogonal coordinate optimization at the back end.The stability and efficiency of map line tracking are improved by means of feature homogenization,inter-frame tracking and map maintenance.3.An inertial visual odometry based on point-line combined features is proposed.This method adopts the visual odometry with point-line combination features,and improves the robustness and positioning accuracy of the SLAM system in texturedeficient scenes by establishing the tightly coupled nonlinear optimization of vision and IMU and the point-line coplanar geometric constraints,and increases the local construction accuracy.The density of the graph.4.Through the data set,the reliability of the method in scene reconstruction and positioning accuracy is verified,and the advantages of this method and the causes of errors compared with point feature SLAM are analyzed.