Research On Compensation Methods For Lost Trajectories Of ORB-SLAM Based On End-to-End Visual Odometry

Simultaneous Localization And Mapping(SLAM)is a technique in which a robot uses the information collected by sensors to establish surrounding environment and estimate its state in the absence of a priori information.SLAM has received much attention in recent years as a robot sensing technique.The ORB-SLAM family(ORB-SLAM,ORB-SLAM2,ORBSLAM3),as monocular SLAM algorithms with excellent comprehensive performance,are the current research hotspots for their improvement.This thesis is dedicated to investigating how to solve the trajectory loss problem of ORBSLAM family.The impact of trajectory loss mainly includes the following two aspects:1.practical application: robot cannot know its position during the trajectory loss period.2.algorithm performance: in the system,trajectory loss means that the lost frames cannot participate in creating map points,and when the camera passes the area where the trajectory loss happened again,the sparse map points in the area may weaken the system performance.This thesis draws on the robustness of end-to-end visual odometry as a trajectory compensation tool and proposes a compensation algorithm for lost trajectories.The main contributions of this paper are as follows:1.An algorithm to compensate for lost trajectories of ORB-SLAM family using end-to-end visual odometry is proposed.The method uses end-to-end visual odometry as a short trajectory compensation tool,which is used to compensate for the lost trajectories in extreme environments,and designs a trajectory adjustment algorithm,including the scale adjustment,optimization and storage of trajectories,to achieve smooth insertion of the compensatory trajectories.It is demonstrated experimentally that the method can estimate more complete trajectories when adjacent submaps are fused,and the accuracy is almost unaffected.2.The reuse method of compensatory trajectory is proposed.This method means that the compensatory trajectory can be processed by the back-end algorithm,such as map point creation and redundant keyframe removal,so that the compensatory trajectory can be used by the subsequent input frames.The method mainly includes the reliability verification of compensatory trajectory,compensatory keyframe decision and back-end processing methods.The experiments prove that the method can effectively determine whether the compensatory trajectory is reliable and select key frames from the trajectory,and these key frames can provide rich map points after the back-end processing,and these map points can be used by the subsequent input frames.