.png){kind=icon}
.png){kind=icon}
Return
Globally Spatial Consistency-Based Maximum Consensus for Efficient Point Cloud Registration
.png){kind=icon}
PRE
AI
DOI：10.1111/phor.70023.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Point cloud registration is a fundamental task in photogrammetry and computer vision to determine the rigid transformation between a pair of overlapping 3D point sets. Most existing methods address the registration problem by establishing correspondences between pairwise point clouds. However, these methods are sensitive to outliers in the correspondences obtained. In this study, we identify the maximum consensus set from initial correspondences to solve the registration problem in which graph clustering and global spatial consistency are considered. Given a set of initial correspondences between a pair of overlapping point clouds, we first construct a graph according to the distance consistency between their initial correspondences, each of which is viewed as a node in the graph. Subsequently, we select representative nodes as seeds and search for their second-order neighbouring nodes to form clusters, leading to candidate sets of the maximum consensus. Subsequently, we identify the optimal maximum consensus set with spatial consistency for the global registration of pairwise point clouds. Finally, we use an adaptive robust loss function and adopt a graduated nonconvexity strategy to iteratively refine the registration. We conducted experiments using two challenging real-world datasets. The experimental results demonstrate that our method achieves accurate performance on pairwise point clouds with a high ratio of outliers among their initial correspondences. In addition, the proposed method was compared with traditional baseline and state-of-the-art methods. The comparison showed higher efficiency and better robustness when our method was used in scenarios involving multiple correspondences. The source code is available at https://github.com/MasterFlashQ/MAGSC.
Keywords:
globally spatial consistency
graph clustering
maximum consensus
point cloud registration
Journal
P
IF:
3.6
Papers:
29
Citations:
0
Organization
No organization information available