Indoor Stereo Visual Localization Based On Point And Line Features And 2D Lidar Mapping

Location and mapping are the two most important problems of autonomous mobile agents in the indoor environment.They are the foundation of other upper level applications.Because the visual sensor can provide abundant environmental information,the research based on Visual SLAM(SLAM)/Visual Odometry(VO)becomes a hot topic.Due to the rich geometrical features of points,lines,and surfaces in the indoor environment,estimating the state of the system by combining information of different dimensions can avoid the loss of features or high uncertainty of the features in the process of estimating the state through a single dimension.Errors in state estimation improve the robustness of the positioning system operation,and then improve the robust operation of the localization system.Because of its high accuracy and stability,laser radar is a common sensor for mapping.At the same time,due to the poor real-time performance of visual SLAM on-line dense construction and the regularity of indoor environment,it is possible to construct a structured map using laser radar.In this thesis,the binocular stereo camera is used to extract ORB and line segment features from environmental images for tracking.Based on ORB-SLAM2,a more robust binocular vision SLAM algorithm is proposed to solve the localization.The 2D laser radar scans the environment and receives the localization results given by the visual SLAM,constructing a 2D line feature global map.The main innovations in this paper are as follows:(1)Proposed and deduced the formula of the 3D space Plucker coordinates of the corresponding straight line from the pixel coordinates of the endpoints of the left and right eyepieces of the binocular stereo camera,and carried out strict mathematical proof.(2)A 3D space parameterization method for horizontal alignment of paired lines segments in binocular camera field is proposed.(3)A robust matching algorithm for 3D spatial line segment feature and 2D image line segment feature is proposed.(4)A new algorithm for rapidly constructing line feature map from 2D lidar data is proposed.The algorithm is based on the similar triangle denoising rule for data denoising.The performance of the vision SLAM algorithm proposed in this paper is tested on the open source dataset.The results prove that the simultaneous use of point and line features can improve the localization accuracy.In this thesis,the visual localization and 2D laser radar mapping integrated system was implemented in engineering.The real environment was tested for mapping performance.The results show that when the linearity of the scanned data is stronger than the one-to-many mapping characteristics it presents,the algorithm with excellent performance.