SLAM Method And Experimental Research For Collaborative Mapping Of Unmanned Aerial Vehicle Clusters

With the development of drone technology and computer vision technology,visual SLAM has become a key research direction in the field of drones.However,it is usually carried out for single drone SLAM,and as the complexity of the actual scene increases,problems such as low mapping efficiency,poor system stability,and limited information collection range may arise in the construction of single drone SLAM.Therefore,this article designs a centralized unmanned aerial vehicle cluster collaborative SLAM framework to study and analyze the communication load and limited processor resources of unmanned aerial vehicle collaborative SLAM.The innovation points and main research work are as follows:This article proposes a threshold based sub terminal optimization method.By using the local map optimization module,the local map of the sub terminal drone is limited within the set threshold range,reducing the computational pressure on the sub terminal drone’s own processor and improving the real-time performance of the system.Through the keyframe optimization module,redundant information in keyframes is detected,eliminating redundant keyframes and improving the quality and transmission efficiency of sub end information.Through experimental comparison,it was found that the local map keyframes of the unmanned aerial vehicle in the subsystem of this system always remain within the set threshold range when the restored point cloud maps are not significantly different.This article proposes an Apriltag label map fusion method.Firstly,a fusion scheme based on scene overlapping maps was designed for experimental simulation of the KITTI dataset.Under special scenario conditions,due to factors such as the flying speed and trajectory of the drone at the sub end,map fusion cannot be performed when the sub end reaches overlapping scenes.Therefore,the Apriltag tag is used to solve the relative pose between drones.Arrange Apriltag tags in special scenarios,and when the drone detects Apriltag tags,force map fusion to improve the overall mapping accuracy and diversity of map fusion of the system.This article proposes a multi drone collaborative SLAM system framework.This framework can simultaneously control multiple drones,perform real-time SLAM mapping,and transmit key frames and map points to the cluster control workstation.Cluster control workstations perform map fusion to build a global map.Compared to single drone SLAM mapping,this system can complete mapping in a faster and more efficient manner in the same scenario.In the Gazebo simulation scenario,compared with the ORB-SLAM system,the point cloud scene restoration is similar,and the number of key frames and map points in this system is less than half of ORB-SLAM.Under the KITTI dataset,the ATE error of this system is 0.91 m,and the ORB-SLAM system error is 0.99 m,with a small difference.This system effectively balances computation,communication pressure,and data collection,reduces the pressure on small end processors,and reduces system communication load.Based on the threshold sub end optimization module and map fusion scheme,good results have been achieved.