LCR-SMPL: Toward Real-Time Human Detection and 3D Reconstruction from a Single RGB Image

Elena Pena-Tapia, Ryo Hachiuma, Antoine Pasquali, Hideo Saito

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper presents a novel method for simultaneous human detection and 3D shape reconstruction from a single RGB image. It offers a low-cost alternative to existing motion capture solutions, allowing to reconstruct realistic human 3D shapes and poses by leveraging the speed of an object-detection based architecture and the extended applicability of a parametric human mesh model. Evaluation results using a synthetic dataset show that our approach is on-par with conventional 3D reconstruction methods in terms of accuracy, and outperforms them in terms of inference speed, particularly in the case of multi-person images.

Original languageEnglish
Title of host publicationAdjunct Proceedings of the 2020 IEEE International Symposium on Mixed and Augmented Reality, ISMAR-Adjunct 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages211-212
Number of pages2
ISBN (Electronic)9781728176758
DOIs
Publication statusPublished - 2020 Nov
Event2020 IEEE International Symposium on Mixed and Augmented Reality, ISMAR-Adjunct 2020 - Virtual, Recife, Brazil
Duration: 2020 Nov 92020 Nov 13

Publication series

NameAdjunct Proceedings of the 2020 IEEE International Symposium on Mixed and Augmented Reality, ISMAR-Adjunct 2020

Conference

Conference2020 IEEE International Symposium on Mixed and Augmented Reality, ISMAR-Adjunct 2020
CountryBrazil
CityVirtual, Recife
Period20/11/920/11/13

Keywords

  • Artificial Intelligence
  • Artificial Intelligence
  • Computer vision
  • Computer vision
  • Computer vision problems
  • Computer vision problems
  • Reconstruction
  • Shape inference

ASJC Scopus subject areas

  • Computer Science Applications
  • Media Technology
  • Modelling and Simulation

Fingerprint Dive into the research topics of 'LCR-SMPL: Toward Real-Time Human Detection and 3D Reconstruction from a Single RGB Image'. Together they form a unique fingerprint.

Cite this