Research On Simultaneous Localization And Map Construction Based On ROS

With the development of mobile robots,Simultaneous Localization and Mapping technology,as the basis of mobile robot application technology,has received extensive research attention.Compared with other sensors,the visual sensor which has the advantages of low cost and richer color and texture information has been more and more widely used.In this paper,a vision sensor is used to research on Simultaneous Localization and Map Construction of Planar Motion Wheeled Robot.This article uses RGB-D camera to collect image information,and implements the overall visual SLAM on the ROS system.The specific experiment content is as follows:Firstly,according to the principle of camera imaging,the camera is calibrated,the internal parameter parameters of the camera are solved,and the basic information is provided for solving the three-dimensional space points through the image.Secondly,based on the motion characteristics of the RGB-D camera plane-moving wheeled robot,in the image processing part of the visual odometer,the contour constraint conditions are established according to the camera motion constraints,which can filter out the mismatched feature point pairs in the color map and the matching point pairs that match the color map correctly but the depth value measurement error is large,and obtain high-quality feature point matching pairs.In the motion calculation part between image frames,the transmission matrix is reduced in dimensionality,and the transmission matrix is linearized and approximated to construct a least squares problem.By dynamically compensating the linearized approximation calculation error,accurate robot camera motion parameters can be obtained.Finally,in the loop detection,the bag-of-words method is used to detect the similarity between key frames,and the camera motion trajectory is optimized through the back-end optimization using graph optimization to eliminate the cumulative error problem.A sparse point cloud map is established by camera pose and landmark points,and an octree map is further established.