Open AccessResearch on the Construction Method of 3D Dense Map Based on Visual Slam
Author(s) -
Tianfang Sun,
Xun Shi
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1871/1/012083
Subject(s) - simultaneous localization and mapping , point cloud , computer science , graph , artificial intelligence , convergence (economics) , computer vision , frame (networking) , set (abstract data type) , pose , robot , mobile robot , theoretical computer science , telecommunications , programming language , economics , economic growth
Aiming at the problem that visual SLAM (Simultaneous Localization and Mapping) cannot satisfy both high-precision positioning and rapid mapping in large scenes, this paper proposes a map construction method. First, when calculating the pose, the method of combining EPNP and graph optimization is used to optimize the pose estimation to improve the accuracy of the pose. Then, the paper uses Lie algebra to define the vertices and edges of the pose graph, which improves the convergence speed of the model. Finally, when constructing a three-dimensional map, a divide-and-conquer map method is proposed, which uses multiple sub-maps that contain each other for point cloud registration and down-sampling filtering, so as to improve the accuracy of map construction. Experimental results show that the algorithm in this paper has good performance. In the TUM test set, the speed can reach 0. 139s/frame, which is about 0.2s higher than the original method, and the accuracy error RMSE value is reduced to 0.055m. The feasibility of the algorithm is proved, and it can meet the accuracy and speed requirements of 3D dense map construction in large scenes.