A Centralized Collaborative Monocular SLAM System with Relocalization

In order to solve the problems of low efficiency and poor robustness of single unmanned system, a centralized multi-robot collaborative SLAM based on monocular camera is proposed in this paper. The system adopts a centralized structure with one server and several clients. Each client runs visual odometry independently and sends the map information and keyframe information to the server. The server performs tasks such as Intra-Map loop closure detection, map matching, map fusion, and map global optimization. To solve the problem of tracking failure caused by the client's fast motion and blurred image information, we keep all the local keyframe information in the client and add relocalization mode in the tracking process. We evaluate the performance of the system on public datasets. Simulation experiments show that in scenes with poor image information, our system is able to re-track to the image using relocalization, which makes our system more stable and robust.