Research On De-noising Algorithm And Registration Algorithm Of Three-dimensional Point Cloud Model Data

With the improvement of modern3D scanning technology,3D point cloud hasbecome an important way of modeling, processing of the point cloud model has become afocus in recent years. In order to reconstruct the object, we first need to get real data of thesurface of object, however, due to limitations and impacts of the measurement equipments,environmental and other such factors, the point cloud data we can obtain are always only apart of surface of the object at each time, and the problems of unreasonable noise,distortion of data and translational and rotational dislocation are inevitable. Therefore,proposing efficient and high robustness de-noising and registration algorithms becomesthe key in the process of the three-dimensional reconstruction.This article study for three-dimensional point cloud model data de-noising andregistration algorithm research and related issues and the main task are as follows:Firstly, aimed at the existing filtering algorithms has the problems of the verticesdrifting and keeping sharp features difficultly, this article give full consideration to theposition of the sampling points of the point cloud model in space, normal vectors andGaussian curvatures information, asymmetrically choose the sampling point neighborhood,proposed an algorithm which uses different filtering strategies on different feature regionsof a point cloud model, anisotropic de-noise the sampling points.Secondly, in order to provide accurate registration better initial values andstreamlined data set, propose a random sampling consensus crude registration algorithmwhich is base on the congruent three elements, this algorithm uses curvature of samplingpoints and the distance from sampling point to the neighborhood centroid to find thecorresponding point set that the three elements are congruent, uses the corresponding setof points instead of the original point cloud registration, then use the SVD method tocalculate transform coordinates.At last, in order to compare with the existing algorithms, to prove the point cloudmodel becoming more smooth while keeping sharp features of the details of the modelbetter when using the filtering algorithm based on curvature features classification byexperiments, and not only has better filtering effect, but also improves the speed and degree of automation of the operation; And the random sampling consensus cruderegistration algorithm base on the congruent three elements improves the efficiency of theregistration, and it has good adaptability, stability and robustness, and can provide goodinitial values for the next precise registration.