Multi-site cloud registration method based on manifold clustering

Abstract

The invention discloses a manifold clustering-based multi-site cloud registration method. The method comprises the following steps of S1, carrying out initial point cloud registration; s2, acquiring an initial centroid; s3, updating the center of mass; s4, updating rigid transformation; s5, repeating the step S3 and the step S4 until an iteration stopping condition is met; s6, calculating an overlapping part proportion and an overlapping part point set; s7, corresponding weights are given to the data points, and rigid transformation is updated until an iteration stopping condition is met; and S8, obtaining a final registration result according to rigid transformation when iteration stops. According to the invention, an initial centroid is selected by using a local neighborhood feature point extraction method, so that the influence on a registration result due to neglect of local geometric features of an object is reduced; the geodesic distance is used as the basis of clustering division, so that the geometric structure of the object can be considered to a great extent; the problem of overlapping parts is solved by using a method of endowing different weights to the data points, and the ghosting problem occurring in the registration result can be optimized.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a multi-site cloud registration method based on manifold clustering. Background technique [0002] With the continuous development of information technology, face recognition technology, automatic driving technology and robot technology have gradually developed, and are more and more closely related to people's lives. In order for these technologies to operate normally, accurate image information needs to be obtained, and image registration technology is particularly important. [0003] With the development of lidar scanning technology, we can quickly obtain point cloud data information in all directions on the surface of three-dimensional objects, and multi-site cloud registration technology can combine data information in all directions to obtain accurate point cloud data. Model. [0004] Point cloud registration technology is to find the corresponding re...

AI Technical Summary

Problems solved by technology

[0005] Usually, the iterative closest point algorithm is used for pairwise registration of two views. The iterative closest point algorithm can only perform pairwise registration of point clouds of two views, and cannot register multiple views. The iterative closest point algorithm There are also problems such as not being able to handle overlapping problems well, slow operation speed, and being greatly affected by initial input parameters.

Images