Research On SLAM Back-end Optimization Based On Improved Graph Optimization

The indoor positioning and navigation technology of indoor service robots has received more and more attention in order to meet the increasing complexity of indoor tasks.Simultaneous Localization and Mapping(SLAM)is a common technique for solving indoor service robot navigation.Filter-based methods and graph-based optimization methods are two common methods for solving SLAM.The SLAM method based on graph optimization(Graph SLAM)has problems such as large data size,low efficiency,and unsuitability for wide application.In this paper,the real-time problem of Graph SLAM is studied.The framework of Graph SLAM system is studied.The back-end error source of Graph SLAM is analyzed.The back-end optimization method of Graph SLAM based on PageRank is proposed to improve the optimization efficiency and real-time performance.The main research contents are as follows:Firstly,research on SLAM based on graph optimization framework.Graph SLAM consists of front-end,closed-loop detection,back-end and construction.These four parts are necessary for the research of Graph SLAM.The front-end mainly completes the pose map construction,and studies the Graph SLAM front-end based on the feature point method,and establishes the SLAM front-end based on the ORB-based feature point extraction and FLANN-based feature point matching.The closed-loop detection is used to detect if it is back to the original position.The closed-loop detection based on the word bag model is studied.The closed-loop detection is completed by establishing a word bag and performing similarity calculation between key frames.The back-end is to optimize the position of the robot and the position of the spatial feature points,and studies the back-end optimization based on BA to reduce the complexity of back-end optimization.The mapping is a map representation of the backend optimization results.Then build the Graph SLAM system to provide a basic framework for back-end optimization research.Secondly,research on back-end optimization error based on RGB-D sensor.Above all,introduce the RGB-D sensor used in SLAM.After that,the source of error in the back-end of the Graph SLAM is analyzed.When the SLAM based on the RGB-D sensor calculates the pose transformation between the two moments of the robot,due to the inherent properties of the sensor,an error occurs,that is,a reprojection error.Based on this error,the Graph SLAM back-end optimization model is established to transform the SLAM problem into a nonlinear least squares optimization problem.The versatility tools for back-end optimization are elaborated.Thirdly,research on back-end optimization algorithm of Graph SLAM based on PageRank.Different nodes have different effects in the same large network node.The pose map constructed in Graph SLAM contains a large number of nodes.Through the research of PageRank algorithm,the PageRank algorithm is used to calculate the importance of the nodes in the pose map.According to the importance of different nodes,the scale of the pose map is reduced,and the back-end optimized data is reduced to improve the real-time performance.Finally,the PageRank-based Graph SLAM back-end optimization algorithm is experimentally verified and the experimental results are analyzed.The experimental results show that the back-end optimization time of Graph SLAM is shortened and the real-time performance is improved.