Point cloud accelerated registration method

Abstract

The invention belongs to the field of image processing, and more particularly relates to a point cloud accelerated registration method, which comprises the following steps: down-sampling original point cloud data, and extracting a first neighborhood point set of points; performing feature extraction on the first neighborhood point set by using a regular icosahedron group to obtain a descriptor with rotation and other denaturation; registering a source point cloud and a target point cloud in the original point cloud data according to the rotation isodenaturation descriptor to obtain a block registration relationship; and carrying out local optimization registration on the block registration relation to complete registration work of the source point cloud and the target point cloud. According to the invention, the speed and accuracy of point cloud registration can be considered.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more particularly relates to a point cloud accelerated registration method. Background technique [0002] With the rapid development of the field of machine vision, good image effects can be obtained by processing point cloud data. [0003] The point cloud is a collection of massive points that express the spatial distribution of the target and the surface characteristics of the target in the same spatial reference system. The points in the point cloud can be expressed by coordinates. [0004] For point clouds of the same object acquired by different devices, point cloud registration is required to obtain more accurate point cloud data. [0005] Point cloud registration is to solve the rotation and translation matrix of the point cloud, and transform the source point cloud into the same coordinate system as the target point cloud; the common registration algorithm is to find the fea...

AI Technical Summary

Problems solved by technology

However, the descriptors in the existing technology do not have good rotation invariance. When the point cloud is rotated, the algorithm cannot obtain the correct point cloud registration as if the rotation did not occur. However, when the local reference coordinate system is used to enhance the rotation invariance , which will affect the density and noise of the point cloud, resulting in poor robustness

[0006] In addition, the current registration algorithm based on deep learning leads to the loss of information of some points in the point cloud during the downsampling process, resulting in inaccurate correspondence between the point pairs obtained through registration, and the registration effect is limited; In the point method, due to the excessive number of points, the network processing takes too long

[0007] In the prior art, a depth image matching method based on point cloud registration first uses deep learning method to obtain rough registration, and then uses ICP algorithm for fine registration. However, the coarse registration and fine registration cannot be combined. Training is a staged process, and the registration accuracy is limited. Therefore, it can be seen that the existing registration methods cannot take into account the accuracy and efficiency of registration.

Images