A Filtering Algorithm for Scattered Point Cloud Based on Curvature Features Classification

Abstract

In order to remove the noise of three-dimensional scattered point cloud, and to smooth point cloud model efiectively, an algorithm which uses difierent flltering strategies on difierent feature regions of a point cloud model is proposed. The algorithm flrstly estimates the normal vectors of a point cloud model by using covariance analysis method, then estimates Gaussian curvatures of the point cloud model by using parabolic fltting method, and uses the average Gaussian curvature in the k-neighborhood as the threshold, to divide the point cloud model into ∞at neighborhood type regions and mutant neighborhood type regions, and, we use improved median flltering algorithm and bilateral flltering algorithm on difierent feature regions, so as to achieve the purpose of smoothing scattered point cloud. Experimental results show that it has a very good smoothing efiect and maintains the detail features of the point cloud model, which has proved the algorithm is simple and efiective.