Keypoint Detection Based on Curvature Grouping and Adaptive Sampling

In the keypoint detection algorithm, the farthest point sampling methods and random sampling methods are usually used to select candidate points, then keypoints are screened out from the neighborhood of the candidate points. In the methods, reproducibility and interpretability of keypoints are relat...

Full description

Saved in:
Bibliographic Details
Main Authors: Bifu Li, Yu Cheng, Weitong Li
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10623631/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the keypoint detection algorithm, the farthest point sampling methods and random sampling methods are usually used to select candidate points, then keypoints are screened out from the neighborhood of the candidate points. In the methods, reproducibility and interpretability of keypoints are relatively poor. To address the problems, this paper introduces point cloud geometric features to build a keypoint detection and description network that makes the keypoints more stable and their geometric properties easier to distinguish. The details of point cloud geometric features include: 1) Grouping the point cloud using curvature, Candidate points and their dilation clusters are found through adaptive random sampling and neighborhood expansion; 2) Aggregating the dilation cluster features by using the channel-position attention mechanism to generate keypoints based on the scores; 3) Jointly generating the descriptors using the dilation cluster features and the attention feature maps. Experiments on ModelNet40 and KITTI datasets show that the method in this paper improves the Repeatability by more than 8% compared to the existing methods, and the improvement is more obvious when the number of keypoints is reduced. Additionally, the inlier rate is enhanced by 8% and 3% on the two datasets, respectively.
ISSN:2169-3536