Research On Points Cloud Registration Based On Coherent Point Drift

Recent years have witnessed rapid development of hardware equipment,which has in turn contributed to increasing collection precision of the original data and popularity of the point-cloud-driven research.So far,point cloud has found wide applications in the field of surveying and mapping,intelligent visual sensing,and virtual reality.The point cloud collection equipment can be used to obtain and process the point cloud data to achieve high-precision and fast-speed three-dimensional modeling.This has been a general research interest.As a basic link of modeling,the point cloud registration can influence the quality of the three-dimensional model with its registration effect.The point cloud registration expectation enables registration of three-dimensional point cloud in different spaces and different poses to the same coordinate system.Based on the coherent point shift algorithm,this paper studies the point cloud registration technology,focuses on discussing two registration algorithms – one is based on the iterative closest point and the other on the Gaussian Mixed Model.To start with,an improved method is proposed for weight setting of the original algorithm.Then,concerning the problem of declining registration speed,an improved plan is put forward.Finally,the registration results are used for fitting,and the spatial model can be obtained.(1)A weight setting method realized through the shift registration algorithm and put forward based on outlier factors.First,outlier factors are used to recognize outlier points in the registration model.After that,the number of outlier points is estimated,and the outlier points in the point cloud are eliminated so as to reduce the influence of outlier points on registration results,weaken the function of the marking mechanism,and optimize the influence of the fixed weight value on the algorithm effect.During the process of registration,the cardinal number of outlier points is used to guide setting of the weighted parameters.The improved method mainly offers a weight estimation algorithm considering lack of registration preciseness resulted from selection of improper weighted parameters by the original algorithm.Through the improved method,the registration failure triggered by a large number of outlier points can be minimized.(2)A new registration method is put forward by combining the weight setting idea with the curvature characteristics.The feature point clouds are extracted based on the similarity of the main curvature,and then adopted for registration.The quadratic equation is employed to realize fitting of the surface.The Gaussian curvature can be given by the surface equation.The main curvature threshold method is used to measure the similarity of point clouds.The similar points reaching the threshold value are extracted for registration.The plan can favorably cope with the degeneration of the original algorithm.Since a larger scale of point clouds can slow down the registration speed,curvature characteristics are employed to reduce the point cloud scale.Meanwhile,the corresponding relationship among point clouds is roughly built.On the one hand,the spatial structure is reserved;on the other hand,the size of point clouds is reduced to achieve the purpose of accelerating registration.(3)The point cloud three-dimensional model can be built in accordance with the improved registration plan.The improved method can be used for registration of point clouds.Then,the point cloud model after smoothing registration can acquire a group of evenly-distributed point cloud data.Finally,through fitting of the surface path based on subdivision of the triangle,a three-dimensional model is built.