Multi-UAV Collaborative Simultaneous Localization And Mapping Focusing On Data Sharing

At present,the unmanned aerial vehicle(UAV)has become an indispensable equipment in the industry and service industry due to its flexible mobility,low cost and no risk of casualties.But in scenarios,such as disaster surveys,bridge detection and geological surveys,drones alone are no longer available.Multi-uav collaboration can expand the scope of the task and reduce the time of the task.Therefore,multi-drone coordination has become a hot issue in the industry.Sharing data among Unmanned Aerial Vehicles(UAVs)is one of key issues in the field of multiple-robot SLAM.In this paper,aiming at problems of sharing data between UAVs during tracking lost andmap fusion,we propose a robust,focusing on Date Sharing Multi-UAV visual SLAM(DSM-SLAM)based on centralized architecture.Work in this paper is mainly reflected in the following three aspects:Firstly,aiming at how to share data among multiple UAVs,this paper proposes a robust,data-sharing collaborative multi-UAV SLAM(focusing on Data Sharing Multi-UAV SLAM,DSM-SLAM),which is characterized by supporting the UAV to exchange data in time through the central server in the event of an abnormal situation;dynamically selecting the map fusion sequence dynamically according to the different conditions of the UAV to construct the map,thereby improving the abnormal processing capability of the SLAM system and The efficiency of data sharing.Secondly,based on the DSM-SLAM framework,the current research status of the existing open source SLAM system is based on the DSM-SLAM framework.This paper proposes a two-step relocation method based on shared local map.It is characterized by the ability to support drones to recover data from other drones that have ever reached this location as soon as possible.The experimental results demonstrate the effectiveness of the two-step relocation method.Thirdly,aiming at the problem of low data sharing efficiency in the process of map fusion,this paper proposes a map fusion method based on hierarchical clustering,which is characterized by dynamic adaptive selection according to the number of matching key frames between local maps.It is most beneficial to the map fusion order of data sharing between drones,thereby improving the efficiency of data sharing among multiple drones.Finally,based on the above research foundation,a collaborative SLAM prototype for data sharing based on ROS is implemented and tested on public data sets.Experimental results on popular public datasets demonstrate the feasibility and effectiveness of the system.