NormNet: Point-wise normal estimation network for three-dimensional point cloud data

Janghun Hyeon, Weonsuk Lee, Joo Hyung Kim, Nakju Doh

Research output: Contribution to journalArticle

Abstract

In this article, a point-wise normal estimation network for three-dimensional point cloud data called NormNet is proposed. We propose the multiscale K-nearest neighbor convolution module for strengthened local feature extraction. With the multiscale K-nearest neighbor convolution module and PointNet-like architecture, we achieved a hybrid of three features: a global feature, a semantic feature from the segmentation network, and a local feature from the multiscale K-nearest neighbor convolution module. Those features, by mutually supporting each other, not only increase the normal estimation performance but also enable the estimation to be robust under severe noise perturbations or point deficiencies. The performance was validated in three different data sets: Synthetic CAD data (ModelNet), RGB-D sensor-based real 3D PCD (S3DIS), and LiDAR sensor-based real 3D PCD that we built and shared.

Original languageEnglish
JournalInternational Journal of Advanced Robotic Systems
Volume16
Issue number4
DOIs
Publication statusPublished - 2019 Jul 1

Fingerprint

Convolution
Sensors
Feature extraction
Computer aided design
Semantics

Keywords

  • 3-D deep learning
  • 3-D indoor LiDAR data set
  • 3-D sensor system
  • Normal estimation
  • robustness

ASJC Scopus subject areas

  • Software
  • Computer Science Applications
  • Artificial Intelligence

Cite this

NormNet : Point-wise normal estimation network for three-dimensional point cloud data. / Hyeon, Janghun; Lee, Weonsuk; Kim, Joo Hyung; Doh, Nakju.

In: International Journal of Advanced Robotic Systems, Vol. 16, No. 4, 01.07.2019.

Research output: Contribution to journalArticle

@article{ba8c020e4ea64d3c9787bd51114a196d,
title = "NormNet: Point-wise normal estimation network for three-dimensional point cloud data",
abstract = "In this article, a point-wise normal estimation network for three-dimensional point cloud data called NormNet is proposed. We propose the multiscale K-nearest neighbor convolution module for strengthened local feature extraction. With the multiscale K-nearest neighbor convolution module and PointNet-like architecture, we achieved a hybrid of three features: a global feature, a semantic feature from the segmentation network, and a local feature from the multiscale K-nearest neighbor convolution module. Those features, by mutually supporting each other, not only increase the normal estimation performance but also enable the estimation to be robust under severe noise perturbations or point deficiencies. The performance was validated in three different data sets: Synthetic CAD data (ModelNet), RGB-D sensor-based real 3D PCD (S3DIS), and LiDAR sensor-based real 3D PCD that we built and shared.",
keywords = "3-D deep learning, 3-D indoor LiDAR data set, 3-D sensor system, Normal estimation, robustness",
author = "Janghun Hyeon and Weonsuk Lee and Kim, {Joo Hyung} and Nakju Doh",
year = "2019",
month = "7",
day = "1",
doi = "10.1177/1729881419857532",
language = "English",
volume = "16",
journal = "International Journal of Advanced Robotic Systems",
issn = "1729-8806",
publisher = "Vienna University of Technology",
number = "4",

}

TY - JOUR

T1 - NormNet

T2 - Point-wise normal estimation network for three-dimensional point cloud data

AU - Hyeon, Janghun

AU - Lee, Weonsuk

AU - Kim, Joo Hyung

AU - Doh, Nakju

PY - 2019/7/1

Y1 - 2019/7/1

N2 - In this article, a point-wise normal estimation network for three-dimensional point cloud data called NormNet is proposed. We propose the multiscale K-nearest neighbor convolution module for strengthened local feature extraction. With the multiscale K-nearest neighbor convolution module and PointNet-like architecture, we achieved a hybrid of three features: a global feature, a semantic feature from the segmentation network, and a local feature from the multiscale K-nearest neighbor convolution module. Those features, by mutually supporting each other, not only increase the normal estimation performance but also enable the estimation to be robust under severe noise perturbations or point deficiencies. The performance was validated in three different data sets: Synthetic CAD data (ModelNet), RGB-D sensor-based real 3D PCD (S3DIS), and LiDAR sensor-based real 3D PCD that we built and shared.

AB - In this article, a point-wise normal estimation network for three-dimensional point cloud data called NormNet is proposed. We propose the multiscale K-nearest neighbor convolution module for strengthened local feature extraction. With the multiscale K-nearest neighbor convolution module and PointNet-like architecture, we achieved a hybrid of three features: a global feature, a semantic feature from the segmentation network, and a local feature from the multiscale K-nearest neighbor convolution module. Those features, by mutually supporting each other, not only increase the normal estimation performance but also enable the estimation to be robust under severe noise perturbations or point deficiencies. The performance was validated in three different data sets: Synthetic CAD data (ModelNet), RGB-D sensor-based real 3D PCD (S3DIS), and LiDAR sensor-based real 3D PCD that we built and shared.

KW - 3-D deep learning

KW - 3-D indoor LiDAR data set

KW - 3-D sensor system

KW - Normal estimation

KW - robustness

UR - http://www.scopus.com/inward/record.url?scp=85068499910&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068499910&partnerID=8YFLogxK

U2 - 10.1177/1729881419857532

DO - 10.1177/1729881419857532

M3 - Article

AN - SCOPUS:85068499910

VL - 16

JO - International Journal of Advanced Robotic Systems

JF - International Journal of Advanced Robotic Systems

SN - 1729-8806

IS - 4

ER -