Research On Point Cloud Registration Algorithm Based On Geometric Features

In recent years,with the vigorous development of hardware and software technologies,point cloud data acquisition and processing technologies related to three-dimensional point cloud have begun to embrace vigorous development.Point cloud data collected by a single sensor can no longer meet development needs.Point cloud registration of point cloud data o btained by multiple sensors can more accurately describe the real three-dimensional world.The purpose of point cloud registration is to transform the scanning models of different data sources into the same coordinate system to obtain a multi-dimensional complete point clou d model.Three-dimensional point cloud registration is the most basic and important part of integrating multi-source point cloud data.This paper first introduces the basic theory and mathematical basis related to point clo ud registration,and then studies the point cloud preprocessing method,point cloud rough re gistration algorithm and point cloud fine registration algorithm.The specific research conte nt is as follows:(1)Cross-source data collected by multiple sensors are not directly applicable to point cloud registration due to the absence of point cloud,inconsistent distribution of point cloud density and different local data modes.Therefore,preprocessing of point cloud is a necessar y step,and differentiated processing of two cross-source point cloud data in the same group should be properly carried out in the preprocessing stage.Different point cloud subsamplin g algorithms and different threshold point cloud denoising algorithms are used to ensure the smoothness of the point cloud while preserving the local and global features of the original point cloud.(2)This paper analyzes the registration accuracy and efficiency of the existing point cloud rough registration algorithm on cross-source point cloud data,proposes to integrate the idea of improved principal component analysis into the rough registration algorithm,and eliminate the non-rigid transformation of point cloud occurrence caused by the registration algorithm based on principal component analysis,and effectively solves the influence of the local missing of point cloud data across source point cloud and uneven dense distribution on registration.Moreover,stable and efficient coarse registration results can be obtained,which provides good initial solutions for subsequent point cloud fine registration algorithms.(3)Aiming at the problems of ICP algorithm for cross-source point cloud registration,such as slow search speed based on exhaustive thought,high sensitivity to the relative posit ion of the registration point cloud,insufficient use of local geometric structure information,and slow convergence caused by too many mismatched points,an ICP algorithm based on g eometric features was proposed.Firstly,the FPFH descriptor is improved to make full use o f the local geometric structure,determine the corresponding point set,and improve the prop ortion of correct point pairs.KD tree is used to replace the search based on exhaustive thou ght,and a certain point cloud topology is established to speed up the search speed between point pairs.The RANSAC algorithm is used to remove the wrong matching point pairs,acc elerate the convergence speed and improve the registration accuracy.Finally,combining the registration results of the principal component rough registration algorithm,a good relative position is provided for ICP algorithm,so as to complete the accurate registration across th e source point cloud.