Research On Global Map Construction Method Based On Multi-Machine Loopback

In recent years,the technology of simultaneous localization and mapping of robots in unknown scenes has attracted a lot of attention and application,but most of the main research focuses on a single robot.Single-machine system has the limitations of limited perception range and easy accumulation of errors,which can not meet the realistic requirements of mission reliability.Compared with single robot,multi-machine system can improve the efficiency of drawing and flexibly adapt to multiple complex scenes.There are some problems in global map construction by multi-robot cooperation.Firstly,how to establish the local data association between multiple robots.Secondly,how to improve the mosaic accuracy of local maps to establish a global map with consistent scale and how to improve the global positioning accuracy of the cluster based on monocular vision.Based on the local positioning and mapping of multiple robots,this paper designs a centralized global map construction scheme based on multi-robot loop method.The research work is as follows:Firstly,the local mapping algorithm of single robot in cluster based on feature point method is studied.ORB method is used to extract features,RANSAC method is used to remove mismatches,and feature matching relationship is established.Motion estimation of pixel pairs is solved by epipolar constraint method,depth of map points is obtained by triangulation method,key frames and map points are screened to build local point cloud map,and octree method is used to build occupied grid map.Secondly,based on the construction of local maps by multiple robots,a local map mosaic method based on multi-robot loop is proposed.Based on ROS topic communication,the map information is shared between the cluster and the main control terminal,and the similarity of image frames shared by multiple robots is described by visual word bag method,that is,the local data association between multiple robots is completed by similar scene recognition.The time consistency test method is used to verify the correctness of multi-machine loop,and sim3 method is used to solve the pose transformation and scale transformation of loop,that is,the similarity transformation during multi-machine loop can improve the mosaic accuracy of local maps and ensure the scale consistency of global mapping;The master terminal adopts the optimization method based on pose map,and the LM method is used in g2 o library to iteratively obtain the optimal pose solution of multi-robots,that is,the global positioning accuracy of the cluster is improved by constructing the global pose map.Finally,a multi-robot experimental platform was built to calibrate the camera,and the calibration results were analyzed.The integrated software based on Qt was developed,and three experimental schemes of short distance,long distance and large range were designed,which qualitatively verified the global mapping effect of the multi-machine system and quantitatively analyzed the global positioning accuracy of the multi-machine system.The experimental results show that the global map construction scheme based on multi-machine loop is effective and reliable.