Multi-scale point cloud feature point detection and matching based on hierarchical b-spline

Abstract

The invention proposes multi-scale point cloud feature point detection and matching based on hierarchical B-spline, which includes the following steps: constructing a surface change metric; dividing the point cloud into two types: potential and non-potential feature points according to the surface change metric , and estimate the noise level of the point cloud according to the non-potential feature points; use the hierarchical B-spline algorithm to fit the potential feature points; record the surface change information during the fitting process to form a curvature scale space, and construct the salient feature points of a stable local coordinate system ; Construct binary feature descriptors according to the distribution information of hierarchical B-spline control points and the curvature size between control points; according to the similarity of descriptors, rely on the topological relationship of feature points to search for corresponding feature points in the model point cloud to achieve feature point matching. The invention can improve the accuracy and repeatability of feature point sampling under the condition of noise and non-uniform sampling, and the binary feature descriptor has higher discrimination and matching success rate compared with the existing feature descriptor.

Description

technical field [0001] The invention belongs to the field of image and visualization, in particular to a multi-scale point cloud feature point detection and matching based on a hierarchical B-spline. Background technique [0002] 3D geometric model has become an important digital medium after sound, image and video. The point cloud model is a geometric model with discrete sampling points as the primitive, and is a natural representation of the 3D geometric model. Feature point detection and matching are an important part of point cloud data processing, and are widely used in multi-view data alignment, 3D model retrieval, surface reconstruction and simplification. However, affected by factors such as sensor noise, missing data, and changes in sampling density, there are still huge challenges in the selection of feature metrics, the repeatability of feature point detection and description, and the accuracy of feature matching. [0003] Feature point extraction and matching c...

AI Technical Summary

Problems solved by technology

Related algorithms include Spin image and its improved algorithm, but since the three-dimensional data is projected onto a plane, it is easy to lose surface information, resulting in poor discrimination of descriptors; Tombari et al. local coordinate system, but the calculation accuracy of the principal curvature direction is easily affected by noise; Mian gives a feature descriptor that is stable under isometric deformation, but the descriptor dimension of the algorithm is very high, which affects the matching efficiency

Images