Optimization solution method for initial registration of real object surface sampling point set

Abstract

The invention provides a point cloud initial registration method combined with a two-dimensional sequence image in order to solve the problem that an existing algorithm is difficult to accurately register point clouds large in initial position difference and small in overlapping area. The method comprises the following steps: acquiring a real object surface grayscale image and three-dimensional data based on a mobile three-dimensional measurement method; positioning the position of the camera in a point cloud local coordinate system according to a perspective projection principle; and obtaining a transformation matrix from the point cloud to a corresponding camera coordinate system, taking the feature points of the grayscale images and the matching points thereof as matching pairs, globally optimizing the camera external parameters by reconstructing the sequence images, and realizing initial registration of the point cloud according to an initial registration formula. According to themethod, point clouds with large initial position differences and small overlapping regions can be correctly registered, and the robustness and the registration efficiency of the registration process can be remarkably improved.

Description

technical field [0001] The invention provides an optimal solution method for the initial registration of a sampling point set on a physical surface, and belongs to the technical field of product reverse engineering. Background technique [0002] 3D digital models are usually used to describe the shape of objects, and are widely used in reverse engineering, machine vision, biomedicine, restoration of cultural relics and other fields. Active optical measurement technology can effectively solve the problem of digital modeling of free-form surface objects, but due to the visibility limitation of a single viewing angle, it is usually necessary to use point cloud registration technology to integrate multiple sets of local point cloud data from different viewing angles into a complete 3D in the digital model. Point cloud registration is an important basis for point cloud post-processing, which directly affects the overall error of the 3D reconstruction of the measured object. [...

AI Technical Summary

Problems solved by technology

Some scholars have proposed a registration method based on probability, assuming that the two point clouds are subject to Gaussian mixture distribution, and constructing a likelihood function, using the maximum expectation algorithm to estimate the optimal rigid transformation parameters, this type of algorithm uses a soft matching strategy, using probability The value measures how well any point matches a potential matching point, but still cannot avoid the dependence on the size of the overlapping region

[0004] To sum up, in the current point cloud registration process, there are too strict requirements on the initial position and overlapping ratio of the sampling point set on the surface of the physical object. It is usually difficult to accurately register point clouds with small overlapping areas. Therefore, the initial position difference Accurate registration of sample point sets on the surface of a large object with a small overlapping area to provide better initial registration parameters has become a technical problem to be solved urgently by those skilled in the art

Images