Sub-map Based Visual SLAM And Navigation Methods For Working Environments

In the complex large-scale indoor working environments,it’s inefficient to explore and construct the map by visual SLAM through one-time.Aiming at large-scale indoor working environments with complex structure,sub-map based multi-session visual SLAM and navigation methods are researched and a general visual mapping and localization module is designed.Based on this module,two functional software are developed for different application scenarios.Firstly,aiming at the characteristics of the large-scale and complicated structure of indoor working environments,a multi-session visual SLAM method based on sub-map is proposed.Based on the ORB-SLAM2,the strategy of automatic creation of sub-map,the calculation of the transformation relationship between sub-maps,and map fusion are mainly researched.Efficient and reliable map construction in complex indoor working environments is achieved.Secondly,aiming at the problem that the traditional visual SLAM lacks a multi-dimensional description of the working environments,a hierarchical data structure is designed to describe the environments and store the map data,including the topological structure of the environments,the metric grid maps of local area,the semantic information of the equipment and the area,the map used for relocalization,etc.,which realizes the multi-level and comprehensive description of the environments.Finally,based on the topological metric mixed map,a hierarchical navigation method is designed.The spatial structure of the environment and the connectivity between regions are reflected by the topological maps,which provides path planning at sub-map level;The accurate measurement information for local areas is reflected by the metric grid maps,which provides accurate path planning results.The task of path planning is completed through the cooperation of them,which greatly improves the success rate and efficiency of path planning in the indoor environments with complex spatial structure.In addition,based on the above research,a general visual mapping and localization module is designed,which is divided into two parts: client and server.The construction and real-time localization in sub-map are realized by the client.The processing,fusion,and management of the map data are realized by the server.Based on this module,the functional software of the wearable assistant system and mobile robot location and navigation is developed for the two practical application scenarios: multi-person collaborative working environment scanning mapping and large-scale indoor mobile robot localization and navigation.Finally,the algorithm and the module are tested in the laboratory environment and the indoor environment of the training substation,and the results verify the effectiveness of the methods.