Research of large-scale offline map management in visual SLAM

This paper presents a novel method of visual simultaneous localization and mapping (SLAM), which is a method of real-time localization and mapping. It is important for a mobile robot to build a map while autonomously navigation. Due to the complexity of the robot work scene, the SLAM method proposed in this paper optimizes map management. It will cost a lot of time and space when a robot long-term works in a same large scene. Therefore, we propose a method in this paper to save a detail map as an offline map in advance. At the same time in order to facilitate the follow-up optimization, the offline map can be divided into several sub-graphs according to the similarity of the scene. Since the segmented offline map has been saved to local system, it can be loaded at any time to localization and obtain the pose of current frame.