Point cloud registration method based on shape constraint

Abstract

The invention discloses a point cloud registration method based on shape constraint, which comprises the following steps of: loading a source point cloud and a template point cloud, selecting a to-be-matched point cloud from the source point cloud, and obtaining a feature descriptor of each point; randomly selecting a plurality of initial points from the to-be-matched point cloud, carrying out repeating for multiple times, and determining a group of initial points with the maximum graph area as a to-be-tested point set; selecting a to-be-matched point from the to-be-measured point set, and calculating distances and included angles between other points in the to-be-measured point set and the to-be-matched point; traversing each point in the template point cloud, searching a reference point, and calculating distances and included angles between other points in the template point cloud and the reference point; screening a pre-selected point set; solving a target point from each pre-selected point set; calculating a transformation matrix between the point set to be measured and the target point set, carrying out coordinate transformation, calculating distance deviation between the matching points, continuously reducing the distance deviation through iteration, and continuously carrying out fine registration on the finally obtained reference transformation matrix by utilizing the final transformation matrix. The constraint on point selection is increased, and the robustness is high.

Description

technical field [0001] The invention relates to the technical field of three-dimensional reconstruction, in particular to a point cloud registration method based on shape constraints. Background technique [0002] 3D point cloud registration technology is an important part of the 3D reconstruction process, and it has a wide range of application prospects in various fields. For example, in the industrial field, it can be used to detect whether there are defects in parts or realize robot grasping work; In the medical industry, it can be used to simulate human organs and find out where the patient's lesions are; in the field of cultural relic restoration, it can be used to restore the full-view point cloud of cultural relics. [0003] In the actual data acquisition process, when using a 3D scanning device to scan the measured object for a single scan, the complete geometric information of the object is often not obtained. Due to the influence of factors such as scanning angle ...

AI Technical Summary

Problems solved by technology

[0004] Point cloud registration is generally divided into point cloud coarse registration and point cloud precise registration. In point cloud registration, if the point cloud coarse registration is more accurate, the robustness of point cloud precise registration will be greatly improved, thus It is not easy to make the algorithm fall into local optimum and improve the convergence speed of the registration algorithm. At present, there are mainly feature-based RANSAC point cloud registration algorithm and Super4PCS-based point cloud registration algorithm in point cloud coarse registration. These two methods The method of randomly selecting points lacks constraints on point selection, which can easily cause disadvantages such as no matching and long calculation time. For example, the feature-based RANSAC point cloud registration algorithm is likely to cause inaccurate matching of corresponding points in the point cloud matching stage. Interference; the Super4PCS-based point cloud registration algorithm takes a long time to calculate in the process of finding the corresponding points of the two point clouds. As the number of point clouds increases, the calculation time doubles and the efficiency is low.

Images