An Algorithm of Anisotropy and Self-adaptive Point Cloud Denoising

摘要

Since the method of isotropic filtering will result in blurring sharp features, a new approach of anisotropic filtering algorithm is proposed, direction and scale of such filter vary according to changes of sharp features'direction and scale. According to the local neighborhood of the point to be processed, the direction and scale of the filter were adaptively computed by covariance analysis. And because the results of covariance analysis are determined by the selection of neighborhood points, in order to adaptively select neighborhood points, a method about the selection of optimal neighborhood radius was also proposed which is based on projection direction. This method adopts projection direction to restrict the selection of neighboring points, thus adaptively get best neighborhood radius related with surface features. Experimental results show that: the algorithm combined with bilateral filtering not only can make point cloud smooth while retaining sharp features but also can adaptively select parameters.