Adaptive markerless three-dimensional point cloud automatic registration method

Abstract

The invention belongs to the point cloud data registration technology in a three-dimensional measurement neighborhood, and particularly relates to an adaptive markerless three-dimensional point cloud automatic registration method. The method comprises steps of search of geometric feature points, search of image feature points, building of a registration algorithm selection model, matching of the geometric feature points based on RANSAC, exclusion of mismatched image feature points by using the RANSAC, solution of a rotation and translation matrix RT by using SVD algorithm, and completion of registration of two point clouds by using the RT matrix finally. The method can be used in measurement situations in which markers can not be attached as object feature points are used for replacing the markers for registration; the method relies on a corresponding feature point to calculate transformation matrix of a multi-view point cloud and does not need to rely on the initial gesture of the point cloud, due to the building of the registration algorithm selection model, the system can adaptively select a proper matching algorithm, thereby realizing stable registration of different measured objects.

Description

[0001] technology neighborhood [0002] The invention relates to an adaptive three-dimensional point cloud automatic splicing method without marker points, which is a method for three-dimensional point cloud data processing and belongs to the point cloud data splicing technology in the field of three-dimensional measurement. Background technique [0003] The three-dimensional measurement technology of surface structured light (refer to literature 1: Li Zhongwei. Research on three-dimensional measurement technology and system of structured light based on digital grating projection [D] [D]. Wuhan: Huazhong University of Science and Technology, 2009) is limited by the single measurement range and requires The measured object is measured multiple times from different orientations to obtain a complete geometric model, among which the automatic stitching of multi-view cloud is the key. [0004] In order to realize the automatic stitching of multi-view point clouds, there are two com...

AI Technical Summary

Problems solved by technology

Among them, the ICP algorithm has high requirements on the initial attitude of the point cloud, and cannot stitch the point cloud with a large difference in the initial position.

The registration algorithm based on geometric features is only suitable for objects with complex surface geometry, and cannot achieve point cloud registration of simple or symmetrical objects

Texture-based registration algorithms are only suitable for objects with rich surface textures, and have poor stability when measuring objects with single textures

[0005] To sum up: the existing 3D point cloud automatic stitching methods have certain limitations and cannot meet the requirements of practical applications.

Images