2D-3D image registration method based on contour coplanar four-point set

Abstract

The invention discloses a 2D-3D image registration method based on a contour coplanar four-point set, and belongs to the technical field of computer vision. The coplanar four-point set is used as a registration primitive. Firstly a point cloud is perspectively projected along multiple visual angles so as to form multiple images, and point cloud images close to the image photographing visual angle are searched; then when the corresponding points of the coplanar four points on the point cloud image contour are searched in the object image contour points, the similar triangles of the triangles formed by the three points are searched firstly and then the approximate position of the fourth point is searched according to affine transformation of the plane; and finally an image and point cloud transformation matrix is calculated by using the homography of the plane, and the optimal transformation matrix is selected according to the registration accuracy corresponding to registration of each transformation matrix so that 2D-3D initial registration can be completed. Compared with the methods in the prior art, the matching corresponding points are searched on the contour so that the method is relatively reliable and simple and great in overall degree of automation of the algorithm.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and mainly relates to the registration of images and point cloud data, specifically a 2D-3D image registration method based on four point sets of coplanar contours, which can be used for rough registration of point clouds and images allow. Background technique [0002] The registration of point cloud and image mainly solves the coloring problem of laser point cloud. Through the registration of point cloud and image, information such as the color in the image can be added to the point cloud, making the point cloud model of the object more realistic and vivid; After point cloud and image registration, it can also assist point cloud registration from different perspectives. [0003] The relatively simple method of point cloud and image registration is to fix the position of the scanner and the camera that takes the image, and take the image at the same time when scanning the object to form ...

AI Technical Summary

Problems solved by technology

Different types of point clouds and images select appropriate feature primitives for registration according to the characteristics of their data, and it is difficult to find a more general matching primitive; most traditional methods also need to know the approximate registration relationship between point clouds and images , project the point cloud according to the initial relationship and then look for the same-named point on the nearby image corresponding to the point cloud feature point, which cannot satisfy complete automation; and most of the feature points in the image are not necessarily the same as the point cloud. are consistent, and it is easy to find wrong matching point pairs

Images