Reconstructing 3D Contour Models of General Scenes from RGB-D Sequences

Weiran Wang, Huijun Di*, Lingxiao Song

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Citation (Scopus)

Abstract

General 3D reconstruction methods use voxels, surfels, or meshes to represent the 3D model of a given scene. These surface-based methods are vulnerable to the loss of boundary details, which affects the completeness of the reconstructed model. In this paper, we focus on the boundary information of the scene and propose a novel method to reconstruct 3D models by using 3D contours extracted from input image sequences. We design a robust frame-to-model contour matching algorithm to solve the problem of finding many-to-many contour correspondences between different frames, and use contour-enhanced optimization to obtain more accurate camera poses. In order to make the reconstructed model more expressive of structural information, we propose a contour fusion algorithm that considers the connections between 3D contours. Compared with other methods which use straight lines or curve segments to reconstruct the scene model, our method can generate a more complete and regular 3D contour model with topological relationship. Experiments on several public datasets demonstrate the effectiveness of our method for both modeling and pose estimation.

Original languageEnglish
Title of host publicationMultiMedia Modeling - 28th International Conference, MMM 2022, Proceedings
EditorsBjörn Þór Jónsson, Cathal Gurrin, Minh-Triet Tran, Duc-Tien Dang-Nguyen, Anita Min-Chun Hu, Binh Huynh Thi Thanh, Benoit Huet
PublisherSpringer Science and Business Media Deutschland GmbH
Pages158-170
Number of pages13
ISBN (Print)9783030983543
DOIs
Publication statusPublished - 2022
Event28th International Conference on MultiMedia Modeling, MMM 2022 - Phu Quoc, Viet Nam
Duration: 6 Jun 202210 Jun 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13142 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference28th International Conference on MultiMedia Modeling, MMM 2022
Country/TerritoryViet Nam
CityPhu Quoc
Period6/06/2210/06/22

Keywords

  • 3D Reconstruction
  • 3D contour model
  • Topological relationship

Fingerprint

Dive into the research topics of 'Reconstructing 3D Contour Models of General Scenes from RGB-D Sequences'. Together they form a unique fingerprint.

Cite this