Point Cloud Classification Using Rotation-Compensated Panoramic Images

View at publisher Summarise with VIRAT
Authors: Yoji Murayama, Junya Hara, Hiroshi Higashi, Yuichi Tanaka
Cited by: 0
References: 29

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In this paper, we propose a point cloud classification (PCC) method based on panoramic images projected from point clouds. A point cloud consists of a number of 3-D coordinates irregularly distributed in space. Since they lack several desirable properties in contrast to standard regularly-structured signals, conventional neural networks (NNs) designed for image processing are not directly applicable. Hence, existing models directly feed point clouds to point-based neural networks: They are more computationally demanding than image-based NNs. To address the issue, we propose a PCC method based on one-shot panoramic image projection. Our classification model comprises three stages. First, the pose of a point cloud is corrected via a geometric transform estimated by a multi-layer perceptron, named R-Net. Second, we convert the point cloud into a panoramic image by projecting it onto a virtual cylindrical screen that encloses the points. Lastly, the panoramic images are fed into a standard NN for images. In the experiment on a standard point cloud dataset ModelNet40, we demonstrate that the proposed method exhibits comparable accuracy to state-of-the-art methods while significantly reducing computational time.

Fields of Study & Subjects

3D Shape Modeling and Analysis Advanced Neural Network Applications Domain Adaptation and Few-Shot Learning

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–6
Publication date
3 February 2026
Language
EN