Application Research Of Anisotropicmultilateral Filter In 3D Point Clouddenoising

With the rapid development of computer science and technology,3D scanning technology has been widely used.However,in the process of obtaining point cloud data,due to the interference of the external environment,the defects of the scanning equipment itself,the human error and other factors lead to the noise in the sampled data.Therefore,it is important to study the denoising and smoothing of point cloud model.Aiming at the problem of over smoothing and local distortion caused by the traditional isotropic point cloud denoising algorithm,this paper makes a deep research on the anisotropic denoising algorithm.Firstly,the sampling point information similarity model was constructed according to the projection of the neighborhood point in the tangent plane of sampling point and the normal of sampling point,then the local neighborhood of the sampling point is selected quantitatively and the effective neighborhood is determined by setting the threshold.The problem of excessive smoothing caused by neighborhood points with large difference of sampling point have the effect on sampling point is solved in high frequency region.Secondly,according to the theory of anisotropic Gauss kernel function,principal component analysis and anisotropic Gauss kernel function are combined,and the distribution properties of eigenvalues and eigenvectors of the surface points,curve points and the corner points is analyzed,and the anisotropic properties are verified by experiments.Thirdly,parametric adaptive and anisotropic Gauss kernel function was designed by taking pseudo inverse matrix of the covariance matrix as the bandwidth matrix,filter main direction and the attenuation velocity of each principal direction can be adjusted adaptively accoding to the local distribution properties,combined it with the bilateral filtering algorithm for scattered point cloud denoising.Finally,the effective neighborhood is combined with the traditional algorithm,and the reliability of the algorithm is verified by experiments.The proposed algorithm is compared with other algorithms to verify the high efficiency and accuracy of the proposedalgorithm.The experimental results show that the proposed algorithm can denoise the point cloud well and preserve original sharp feature at the same time.