Research On Visual Mapping And Localization For Mobile Robot

With the continuous advancement of science and technology,intelligent autonomous mobile robots have gradually become an important research direction and research hotspot in the field of robotics research.Mobile robots mainly face four technical problems:perception,positioning,planning,and control.Among them,robot positioning is a hot research content in the field of robot autonomous movement and work.This paper takes visual positioning algorithm as the main research object,mainly for visual mapping,optimization,merging,compression,and based on known map positioning algorithms to improve the accuracy,robustness and positioning stability of visual maps The main contents of this paper are as follows:1.Research on offline map optimization algorithm.Perform offline processing on the map constructed by real-time SLAM to reduce the cumulative error and observation error of the map.A hierarchical map optimization algorithm is proposed,and the optimization is divided into three parts:closed-loop detection,pose graph optimization,and global BA.In the closed-loop detection part,a closed-loop detection algorithm based on state compression and common-view filtering is proposed to improve the speed and accuracy of closed-loop detection;in the pose graph optimization part,two different pose graph optimization models are analyzed;the global BA part uses the global reprojection error,and chisquare test to remove outliers.Improve the accuracy of the visual map through these three-part layer optimizations.2.Multi-map merging and map compression algorithm research.In the part of multi-map merging,a multi-map merging model is established,a map merging algorithm of PCM and RANSAC is proposed,and the rotation summation algorithm is analyzed.The map compression algorithm analyzes the key frame selection algorithm,and proposes a map point screening algorithm based on ILP and QLP,which reduces the redundant information of the visual map.3.Fusion of known map positioning algorithm research.A visual positioning framework integrating the map side and the positioning side is proposed.For the scale and multi-frame observation problems,the map end uses the gDLS algorithm to solve the pose,which improves the accuracy of the pose solution.The positioning end integrates vision and IMU for local pose estimation,and integrates global information on the map end to update the global pose of the robot in real time.4.Experimental verification.Carry out the construction of mobile robot hardware and software platform,design and carry out the verification experiments of the accuracy and robustness of the mapping and positioning algorithms to verify the feasibility and effectiveness of the relevant algorithms.