Guided Filter Simplification Method for Noisy Point Cloud Data

In recent years, 3D scanning technology has become increasingly popular, with large amounts of raw point cloud data being captured at a reduced expense in people's daily lives. However, the raw point clouds gathered by the sensors collecting the data are inevitably contaminated with various kinds of noise, leading to disadvantages in practical applications. During data processing with point clouds, the volume of noisy data sets is the main burden. In this context, simplification about the original point clouds is crucial. The goal concerning noisy point cloud simplification denotes reducing the amount of data while preserving the geometric characteristics of the data. The guided filter is always considered as a feature-preserving smoothing operator with temporal validity for 2D images. In this paper, the guided filter is applied to adjust the location for the point cloud, especially the noise points, to enhance the features and geometric details. First, add Gaussian noise to our input point cloud. After that, adjust the noisy data with the guided filter. Then, a significant factor named importance should be calculated. The importance of one point is quantified according to its proximity and shows well its contribution to the geometric expression of the local plane. After normalization, if a point's importance reaches a certain level, it will replace its neighborhood. We compare our method with those commonly used in Point Cloud Library. At the same time, we apply this method to the point cloud collected during the aluminum casting process. The proposed algorithm's effectiveness is shown by the simplified outcomes of several noisy point cloud datasets.