Simultaneous Localization And Mapping Algorithm Using Point And Line Features

The intelligent robot is one of the emerging industries of artificial intelligence.Both the fundermental research and innovative applicatons of robots have received considerable attention in recent years.An intelligent robot should have the ability to understand complex dynamic scenes,as well as the ability to move autonomously in a complex environment.Simultaneous localization and mapping algorithm(SLAM)is the core technique to realize the autonomous movement of robots.The SLAM technique can estimate the motion of a robot and establish a model against an unknown environment,and it is,therefore,used in many fields,such as augmented reality,autonomous vehicles,and unmanned aerial vehicles,and so forth.If input information of the SLAM system is visual information collected from a camera,the system is so-called the visual SLAM.In some artificial environments,the stability of the visual SLAM system is affected by textureless scenes and illumination changes.Additionally,the SLAM algorithm should have the ability to recognize and understand high-level information in the environment.Grid maps are extensively used in robot navigation.Segmenting free spaces of the grid map can obtain semantic information for robot global localization.In order to enhance the processing ability of the visual SLAM system for special scenes and extract the semantic information in the grid map,this thesis devotes to developing a stereo SLAM algorithm using point with line features and a graph-based grid map segmentation algorithm for robot global localization.The major contributions of this thesis are summarized as follows:(1)Proposing a line segment matching method based on point and line features coupling.In this thesis,the Grid-based Motion Statistics method is used to eliminate the mismatch of point features,and then,a larger number of correct matches can be obtained.Line segment features do not have global constraints.The multi-level homography matrices are estimated by the matching of point features to establish geometric constraints of line segment features.A sparse minimization model is established according to the geometric relationship between different features in the image.After solving the optimization problem,the line segment matching of the point and line features coupling is realized.The accuracy and real-time performance of the proposed method is examined by open source data.(2)Developing a stereo visual SLAM algorithm by using point and line features.In order to enhance the processing ability of the visual SLAM system for special scenes,the distribution of points and line features on the image is considered in this work to strengthen the relation between the two features and improve the feature matching of PL-SLAM system.In the stereo matching,the geometric constraints between line segments features are established in accordance to the stereo camera model for line segment feature matching.In the frame tracking,the camera motion model is used to predict the pose of the current frame.Subsequently,the projection position of the spatial point or the line segment at the current frame is determined.The feature matching is,then,implemented based on geometric constraints.When a new key frame is inserted,the landmarks in the local map are associated with the features of the new key frame based on the estimated pose.The test results show that the stereo vision SLAM system operates stably in the case of low-texture scenes and varying illumination.(3)Proposing an approach to graph-based grid map segmentation for robot global localization.The free space of the grid map represents the region that the robot can travel through.By the clustering algorithm,the grid map is divided into multiple clusters.The clusters of the map are represented by a graph,and then,merged into different regions.The segmentation regions show all the assessable areas,such as corridors and rooms.The experimental results from the proposed method are more consistent to the truth segmentation as compared to other methods.The map segmentation method proposed to provide a solution for robot global localization.