Connectivity-based convolutional neural network for classifying point clouds

The acquisition of point clouds with a 3D scanner often yields large-scale, irregular, and unordered raw data, which hinders the classification of objects from these data. Some studies have introduced a method of applying the point clouds to convolutional neural networks (CNNs). This is achieved after preprocessing the volume metrics or multi-view images. However, this method has a limited resolution and a low classification accuracy in comparison to heavy computation in object classification. In this paper, DenX-Conv is proposed to improve the accuracy of object classification while securing the connectivity of points from the raw point cloud. DenX-Conv can extract effective local geometric features by finding the neighbor connectivity based on the geometric topology information of the points. In addition, stable feature learning is made possible by applying a densely connected network to PointCNN's chi-Conv. Application of DenX-Conv to the ModelNet40 dataset resulted in a classification accuracy of 92.5%. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

DenX-Conv is proposed for improving object classification accuracy while maintaining connectivity of points in raw point clouds. By extracting effective local geometric features and applying a densely connected network, a classification accuracy of 92.5% was achieved on the ModelNet40 dataset.