Convolutional neural network in the classifying problem of point clouds in three-dimensional space

Abstract

Most 3D acquisition devices create point clouds in 3D space. The use of convolutional neural networks (CNN) in image analysis shows the advantage of neural network algorithms over classical algorithms for some types of problems. The success of neural networks in analyzing 2D images has led to adaptation of neural networks for processing of 3D data. Point clouds by their nature do not explicitly contain information about the geometry of an object; therefore, methods that restore the geometric characteristics of a point cloud can enhance the capabilities of neural networks. In the paper, we propose a convolutional neural network for processing point clouds in three-dimensional space. The network classifies objects defined by point clouds. The proposed approach does not depend on the choice of numbering of points in the cloud. With the help of computer simulation, we show the performance of the proposed network on ModelNet40 database.