基于ORB-SLAM2算法的改进关键帧选择与稠密地图OA

Aiming at the problem of poor real-time performance and accuracy of the key frames selection method of the origi-nal ORB-SLAM2 algorithm,this paper proposes to improve the key frames selection method on the basis of the ORB-SLAM2 algo-rithm,while the traditional ORB-SLAM2 algorithm does not build a dense map and cannot meet the three-dimensional path plan-ning of the robot,so this paper proposes to add a dense point cloud map and a 3D Octomap construction thread on the basis of the original algorithm.The improved algorithm first limits the relative change between frames on the basis of the original algorithm to en-sure the matching accuracy between adjacent keyframes,then makes a comparison based on inliers,and finally sets the minimum feature change threshold to further process the keyframes,and constructs dense point cloud maps and 3D Octomap after eliminating redundant keyframes.Experimental results show that the positioning accuracy of the ORB-SLAM2 algorithm with improved key-frame selection on the TUM dataset is improved by about 24%on average compared with the original ORB-SLAM2 algorithm,and it can effectively construct dense point cloud maps and 3D Octomap in real time to meet the 3D path planning of robots.