Localization And Mapping Of Mobile Robot In A Large Range Dynamic Indoor Environment Based On Graph Optimization

In order to allow the robot to move autonomously in the working environment,the two basic functions of positioning and mapping the environment are indispensable.After years of development,a relatively mature framework for Visual Simultaneous Localization and Mapping(VSLAM)has been developed,which can provide basic environment awareness for robots.The front end of SLAM system framework is the data entry of the whole framework,and its main part is visual odometer,is mainly responsible for the real-time estimates of the robot pose and space points,so the robustness of visual odometer will affect the stability of the whole system,so in this paper,the visual feature extraction and matching process of odometer is studied.In addition,when faced with a relatively large range of complex external environment,due to the limitations of robot hardware and algorithm performance,the mapping and management of existing SLAM algorithms for a large range of environment still need to be improved.Moreover,for realtime operation,most current SLAM algorithms draw sparse feature point maps,which can only be used for basic positioning and do not support advanced functions such as navigation,path planning or scene reconstruction.Aiming at the above problems,some research and improvements on visual SLAM systems have been done from the following points:(1)Firstly,in this paper,the development history of graph-optimized SLAM and the classical framework of graph-optimized visual SLAM system are introduced.The frontend visual odometer,loop detection,optimal-based back-end,mapping based on feature points and point-cloud-based 3D dense reconstruction modules in the model are respectively introduced.(2)Secondly,the improved feature matching method based on the distribution of feature points in image between adjacent frames based on ORB algorithm is introduced.The advantages of ORB(Oriented Fast and Rotated Brief)algorithm are demonstrated by experimental comparison with some current mainstream feature detection algorithms.When extracting ORB features in a single frame image,the original ORB feature extraction algorithm only extracts and describes each individual ORB feature point in the image,without using the distribution information of ORB features in the image.In interframe matching,it is easy to mismatch two feature points with similar descriptors.Therefore,the original ORB algorithm was improved and the distribution position of ORB feature points in the image was added as auxiliary information to improve the efficiency and accuracy of feature matching,and then improve the efficiency and accuracy of the visual odometer were improved.(3)Then,on the basis of in-depth study of ORB?SLAM2,an open source algorithm based on graph optimization,we further focus on saving and updating maps for a wide range of scenarios and explain the detail process of map saving in a wide range of scenarios.At the same time,aiming at the dynamic changing part of the map when drawing the map,how to update the changed part of the map to the existing map is discussed.(4)Finally,a visual SLAM System is designed and implemented on the basis of ORB?SLAM2 algorithm framework by using the existing Robot hardware platform and ROS(Open Source Robot Operating System).The system is capable of real-time 3D point cloud dense reconstruction of the environment,environment map preservation and dynamic part update,map reloading and robot relocation.The actual performance of the system in different scenarios is verified through actual field tests.The final results show that the SLAM system designed in this paper can successfully save and update the map in a large range of scenes,and can also reload the saved map and realize the relocation function on the map.