TY - JOUR
T1 - MFIPC: Point Cloud Registration Algorithm via Multi-Feature Fusion and Interval Pairing Consistency
AU - Wei , Xin
AU - Tu , Bing
AU - Chen , Siyuan
AU - Zhou , Jiadong
JO - Journal of Information and Computing Science
VL - 2
SP - 171
EP - 190
PY - 2025
DA - 2025/05
SN - 19
DO - http://doi.org/10.4208/JICS-2024-010
UR - https://global-sci.org/intro/article_detail/jics/24076.html
KW - Point cloud registration method, Multi-feature fusion, Interval pairing consistency.
AB - <p style="text-align: justify;">Inspired by the Fast point Feature Histogram (FPFH) feature extraction
algorithm, this paper proposes a new 3D point cloud registration method, MFIPC
(Multi-feature Fusion and Interval Pairing Consistency). The method uses feature fusion
and interval pairwise consistency to improve the registration accuracy. In the MFIPC
framework, the point cloud is first downsampled to optimize computational efficiency
and expand the analysis domain. Then, clustering algorithm using local directional
centrality (CDC) classification algorithm is used to calculate the DCM (directional
centrality measure) value of each point. The Gaussian curvature values of the points
are calculated at the same time, and these eigenvalues are fused. To further refine
the registration process, the range between the minimum and maximum eigenvalues is
divided into several equal intervals and sorted in ascending order. A sorting algorithm
is used to assign each eigenvalue to a corresponding interval. For the global point
cloud computing step, after the operation is completed, the number of points in each
interval and its proportion are calculated. The program processes both point clouds
in order to analyze their interval percentage. This algorithm significantly improves the
robustness of MFIPC in establishing point correspondence. To verify the effectiveness
of MFIPC for 3D point cloud registration, we conducted extensive testing on various
datasets, including 3DMatch, RESSO, ModelNet40, Stanford Rabbit, and Dragon. The
experimental results show that the algorithm has high efficiency, good consistency of
point cloud, significantly reduced registration errors, low error and high registration
accuracy</p>