Body-part templates for recovery of 2D human poses under occlusion

Ronald Walter Poppe; Mannes Poel

doi:10.1007/978-3-540-70517-8_28

Body-part templates for recovery of 2D human poses under occlusion

Ronald Walter Poppe, Mannes Poel

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

4 Citations (Scopus)

237 Downloads (Pure)

Abstract

Detection of humans and estimation of their 2D poses from a single image are challenging tasks. This is especially true when part of the observation is occluded. However, given a limited class of movements, poses can be recovered given the visible body-parts. To this end, we propose a novel template representation where the body is divided into five body-parts. Given a match, we not only estimate the joints in the body-part, but all joints in the body. Quantitative evaluation on a HumanEva walking sequence shows mean 2D errors of approximately 27.5 pixels. For simulated occlusion of the head and arms, similar results are obtained while occlusion of the legs increases this error by 6 pixels.

Original language	Undefined
Title of host publication	Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO)
Editors	F.J. Perales, R.B. Fisher
Place of Publication	Berlin
Publisher	Springer
Pages	289-298
Number of pages	10
ISBN (Print)	978-3-540-70516-1
DOIs	https://doi.org/10.1007/978-3-540-70517-8_28
Publication status	Published - 27 Jun 2008
Event	International Conference on Articulated Motion and Deformable Objects, AMDO - Port d’Andratx, Spain Duration: 9 Jul 2008 → 11 Jul 2008

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer Verlag
Number	274
Volume	5098
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	International Conference on Articulated Motion and Deformable Objects, AMDO
Period	9/07/08 → 11/07/08
Other	July 9-11, 2008

Keywords

EWI-12886
EC Grant Agreement nr.: FP6/033812
METIS-251011
IR-64813

Access to Document

10.1007/978-3-540-70517-8_28

Cite this

Poppe, R. W., & Poel, M. (2008). Body-part templates for recovery of 2D human poses under occlusion. In F. J. Perales, & R. B. Fisher (Eds.), Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO) (pp. 289-298). Article 10.1007/978-3-540-70517-8_28 (Lecture Notes in Computer Science; Vol. 5098, No. 274). Springer. https://doi.org/10.1007/978-3-540-70517-8_28

@inproceedings{ef24193324a749bcbf72676556687a18,

title = "Body-part templates for recovery of 2D human poses under occlusion",

abstract = "Detection of humans and estimation of their 2D poses from a single image are challenging tasks. This is especially true when part of the observation is occluded. However, given a limited class of movements, poses can be recovered given the visible body-parts. To this end, we propose a novel template representation where the body is divided into five body-parts. Given a match, we not only estimate the joints in the body-part, but all joints in the body. Quantitative evaluation on a HumanEva walking sequence shows mean 2D errors of approximately 27.5 pixels. For simulated occlusion of the head and arms, similar results are obtained while occlusion of the legs increases this error by 6 pixels.",

keywords = "EWI-12886, EC Grant Agreement nr.: FP6/033812, METIS-251011, IR-64813",

author = "Poppe, {Ronald Walter} and Mannes Poel",

note = "10.1007/978-3-540-70517-8_28 ; International Conference on Articulated Motion and Deformable Objects, AMDO ; Conference date: 09-07-2008 Through 11-07-2008",

year = "2008",

month = jun,

day = "27",

doi = "10.1007/978-3-540-70517-8_28",

language = "Undefined",

isbn = "978-3-540-70516-1",

series = "Lecture Notes in Computer Science",

publisher = "Springer",

number = "274",

pages = "289--298",

editor = "F.J. Perales and R.B. Fisher",

booktitle = "Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO)",

address = "Germany",

}

Poppe, RW & Poel, M 2008, Body-part templates for recovery of 2D human poses under occlusion. in FJ Perales & RB Fisher (eds), Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO)., 10.1007/978-3-540-70517-8_28, Lecture Notes in Computer Science, no. 274, vol. 5098, Springer, Berlin, pp. 289-298, International Conference on Articulated Motion and Deformable Objects, AMDO, 9/07/08. https://doi.org/10.1007/978-3-540-70517-8_28

Body-part templates for recovery of 2D human poses under occlusion. / Poppe, Ronald Walter; Poel, Mannes.
Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO). ed. / F.J. Perales; R.B. Fisher. Berlin: Springer, 2008. p. 289-298 10.1007/978-3-540-70517-8_28 (Lecture Notes in Computer Science; Vol. 5098, No. 274).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Body-part templates for recovery of 2D human poses under occlusion

AU - Poppe, Ronald Walter

AU - Poel, Mannes

N1 - 10.1007/978-3-540-70517-8_28

PY - 2008/6/27

Y1 - 2008/6/27

N2 - Detection of humans and estimation of their 2D poses from a single image are challenging tasks. This is especially true when part of the observation is occluded. However, given a limited class of movements, poses can be recovered given the visible body-parts. To this end, we propose a novel template representation where the body is divided into five body-parts. Given a match, we not only estimate the joints in the body-part, but all joints in the body. Quantitative evaluation on a HumanEva walking sequence shows mean 2D errors of approximately 27.5 pixels. For simulated occlusion of the head and arms, similar results are obtained while occlusion of the legs increases this error by 6 pixels.

AB - Detection of humans and estimation of their 2D poses from a single image are challenging tasks. This is especially true when part of the observation is occluded. However, given a limited class of movements, poses can be recovered given the visible body-parts. To this end, we propose a novel template representation where the body is divided into five body-parts. Given a match, we not only estimate the joints in the body-part, but all joints in the body. Quantitative evaluation on a HumanEva walking sequence shows mean 2D errors of approximately 27.5 pixels. For simulated occlusion of the head and arms, similar results are obtained while occlusion of the legs increases this error by 6 pixels.

KW - EWI-12886

KW - EC Grant Agreement nr.: FP6/033812

KW - METIS-251011

KW - IR-64813

U2 - 10.1007/978-3-540-70517-8_28

DO - 10.1007/978-3-540-70517-8_28

M3 - Conference contribution

SN - 978-3-540-70516-1

T3 - Lecture Notes in Computer Science

SP - 289

EP - 298

BT - Proceedings of the International Conference on Articulated Motion and Deformable Objects (AMDO)

A2 - Perales, F.J.

A2 - Fisher, R.B.

PB - Springer

CY - Berlin

T2 - International Conference on Articulated Motion and Deformable Objects, AMDO

Y2 - 9 July 2008 through 11 July 2008

ER -