.png){kind=icon}
.png){kind=icon}
Return
Redundant same sequence point cloud registration
.png){kind=icon}
PRE
AI
DOI：10.1007/s00371-023-03203-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Many point cloud registration methods rely on establishing correspondence pairs in order to solve the registration problem. However, their performance significantly deteriorates when the given point clouds contain only a limited number of inliers. In this regard, we propose an innovative approach called Redundant Same Sequence point cloud Registration (RSSR) that effectively addresses the challenge of registering point clouds with a limited number of inliers. RSSR achieves this by reconstructing the correspondence relation constraint to allow the discovery of more potential correspondence pairs. It allows one point to correspond with an arbitrary number of points and then generates redundant same sequence point clouds based on the correspondence relation. It explores more potential correspondence pairs at the cost of introducing more erroneous correspondence pairs. Finally, for the introduced erroneous correspondence pairs, we design a rigid transformation solver to register the redundant same sequence point clouds. Our experiments demonstrate that while the performance of other methods significantly drops as the number of inliers decreases, our method maintains stable performance. The codes are available in (https://github.com/ChenPointCloud/RSSR).
Keywords:
Point cloud registration
3D point matching
Computer vision