Primitive fitting using deep geometric segmentation

Duanshun Li, Chen Feng

Research output: Contribution to conferencePaper

Abstract

To identify and fit geometric primitives (e.g., planes, spheres, cylinders, cones) in a noisy point cloud is a challenging yet beneficial task for fields such as reverse engineering and as-built BIM. As a multi-model multi-instance fitting problem, it has been tackled with different approaches including RANSAC, which however often fit inferior models in practice with noisy inputs of cluttered scenes. Inspired by the corresponding human recognition process, and benefiting from the recent advancements in image semantic segmentation using deep neural networks, we propose BAGSFit as a new framework addressing this problem. Firstly, through a fully convolutional neural network, the input point cloud is point-wisely segmented into multiple classes divided by jointly detected instance boundaries without any geometric fitting. Thus, segments can serve as primitive hypotheses with a probability estimation of associating primitive classes. Finally, all hypotheses are sent through a geometric verification to correct any misclassification by fitting primitives respectively. We performed training using simulated range images and tested it with both simulated and real-world point clouds. Quantitative and qualitative experiments demonstrated the superiority of BAGSFit.

Original languageEnglish (US)
Pages780-787
Number of pages8
StatePublished - Jan 1 2019
Event36th International Symposium on Automation and Robotics in Construction, ISARC 2019 - Banff, Canada
Duration: May 21 2019May 24 2019

Conference

Conference36th International Symposium on Automation and Robotics in Construction, ISARC 2019
CountryCanada
CityBanff
Period5/21/195/24/19

Fingerprint

Reverse engineering
Cones
Semantics
Neural networks
Experiments
Deep neural networks

Keywords

  • As-Built BIM
  • Geometric Segmentation
  • Primitive Fitting

ASJC Scopus subject areas

  • Artificial Intelligence
  • Building and Construction
  • Human-Computer Interaction

Cite this

Li, D., & Feng, C. (2019). Primitive fitting using deep geometric segmentation. 780-787. Paper presented at 36th International Symposium on Automation and Robotics in Construction, ISARC 2019, Banff, Canada.

Primitive fitting using deep geometric segmentation. / Li, Duanshun; Feng, Chen.

2019. 780-787 Paper presented at 36th International Symposium on Automation and Robotics in Construction, ISARC 2019, Banff, Canada.

Research output: Contribution to conferencePaper

Li, D & Feng, C 2019, 'Primitive fitting using deep geometric segmentation' Paper presented at 36th International Symposium on Automation and Robotics in Construction, ISARC 2019, Banff, Canada, 5/21/19 - 5/24/19, pp. 780-787.
Li D, Feng C. Primitive fitting using deep geometric segmentation. 2019. Paper presented at 36th International Symposium on Automation and Robotics in Construction, ISARC 2019, Banff, Canada.
Li, Duanshun ; Feng, Chen. / Primitive fitting using deep geometric segmentation. Paper presented at 36th International Symposium on Automation and Robotics in Construction, ISARC 2019, Banff, Canada.8 p.
@conference{442e2de5d1f240a3b696026f1f62e220,
title = "Primitive fitting using deep geometric segmentation",
abstract = "To identify and fit geometric primitives (e.g., planes, spheres, cylinders, cones) in a noisy point cloud is a challenging yet beneficial task for fields such as reverse engineering and as-built BIM. As a multi-model multi-instance fitting problem, it has been tackled with different approaches including RANSAC, which however often fit inferior models in practice with noisy inputs of cluttered scenes. Inspired by the corresponding human recognition process, and benefiting from the recent advancements in image semantic segmentation using deep neural networks, we propose BAGSFit as a new framework addressing this problem. Firstly, through a fully convolutional neural network, the input point cloud is point-wisely segmented into multiple classes divided by jointly detected instance boundaries without any geometric fitting. Thus, segments can serve as primitive hypotheses with a probability estimation of associating primitive classes. Finally, all hypotheses are sent through a geometric verification to correct any misclassification by fitting primitives respectively. We performed training using simulated range images and tested it with both simulated and real-world point clouds. Quantitative and qualitative experiments demonstrated the superiority of BAGSFit.",
keywords = "As-Built BIM, Geometric Segmentation, Primitive Fitting",
author = "Duanshun Li and Chen Feng",
year = "2019",
month = "1",
day = "1",
language = "English (US)",
pages = "780--787",
note = "36th International Symposium on Automation and Robotics in Construction, ISARC 2019 ; Conference date: 21-05-2019 Through 24-05-2019",

}

TY - CONF

T1 - Primitive fitting using deep geometric segmentation

AU - Li, Duanshun

AU - Feng, Chen

PY - 2019/1/1

Y1 - 2019/1/1

N2 - To identify and fit geometric primitives (e.g., planes, spheres, cylinders, cones) in a noisy point cloud is a challenging yet beneficial task for fields such as reverse engineering and as-built BIM. As a multi-model multi-instance fitting problem, it has been tackled with different approaches including RANSAC, which however often fit inferior models in practice with noisy inputs of cluttered scenes. Inspired by the corresponding human recognition process, and benefiting from the recent advancements in image semantic segmentation using deep neural networks, we propose BAGSFit as a new framework addressing this problem. Firstly, through a fully convolutional neural network, the input point cloud is point-wisely segmented into multiple classes divided by jointly detected instance boundaries without any geometric fitting. Thus, segments can serve as primitive hypotheses with a probability estimation of associating primitive classes. Finally, all hypotheses are sent through a geometric verification to correct any misclassification by fitting primitives respectively. We performed training using simulated range images and tested it with both simulated and real-world point clouds. Quantitative and qualitative experiments demonstrated the superiority of BAGSFit.

AB - To identify and fit geometric primitives (e.g., planes, spheres, cylinders, cones) in a noisy point cloud is a challenging yet beneficial task for fields such as reverse engineering and as-built BIM. As a multi-model multi-instance fitting problem, it has been tackled with different approaches including RANSAC, which however often fit inferior models in practice with noisy inputs of cluttered scenes. Inspired by the corresponding human recognition process, and benefiting from the recent advancements in image semantic segmentation using deep neural networks, we propose BAGSFit as a new framework addressing this problem. Firstly, through a fully convolutional neural network, the input point cloud is point-wisely segmented into multiple classes divided by jointly detected instance boundaries without any geometric fitting. Thus, segments can serve as primitive hypotheses with a probability estimation of associating primitive classes. Finally, all hypotheses are sent through a geometric verification to correct any misclassification by fitting primitives respectively. We performed training using simulated range images and tested it with both simulated and real-world point clouds. Quantitative and qualitative experiments demonstrated the superiority of BAGSFit.

KW - As-Built BIM

KW - Geometric Segmentation

KW - Primitive Fitting

UR - http://www.scopus.com/inward/record.url?scp=85071517079&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85071517079&partnerID=8YFLogxK

M3 - Paper

SP - 780

EP - 787

ER -