DEVELOPMENT OF ALGORITHMS FOR POINT CLOUD FILTRATION

Abstract

Today, point clouds are becoming an increasingly common digital representation of realworld objects. However, the raw point clouds obtained by terrestrial laser scanning (or other methods, e. g., photogrammetry or low-cost sensors) are often noisy with many outliers. Thus, it is necessary to remove the noise and outliers from the point clouds before further processing while preserving the elements of the measured objects in high detail. Moreover, in the case of model creation from point clouds using basic geometric shapes (e.g., planes, spheres, cylinders, etc.), one of the most important processing steps is the segmentation of these shapes. Therefore, filtration of unrelated parts of the point cloud can increase the efficiency of processing. In this paper, two algorithms for point cloud filtration are developed, which can be performed based on the local point density and the local normal variation in the surrounding of the selected point. The algorithms were implemented as a standalone application in MATLAB software. The paper's final part describes the experimental testing of the proposed algorithms on several point clouds with various densities, complexity, and different levels of noise and outliers.