Monocular Image 3D Human Pose Estimation under Self-Occlusion

Ibrahim Radwan, Abhinav Dhall, Roland GOECKE

Research output: A Conference proceeding or a Chapter in BookConference contribution

34 Citations (Scopus)
1 Downloads (Pure)

Abstract

In this paper, an automatic approach for 3D pose reconstruction from a single image is proposed. The presence of human body articulation, hallucinated parts and cluttered background leads to ambiguity during the pose inference, which makes the problem non-trivial. Researchers have explored various methods based on motion and shading in order to reduce the ambiguity and reconstruct the 3D pose. The key idea of our algorithm is to impose two types of constraints: kinematic constraints and orientation constraints. The kinematic constraints are imposed by projecting a 3D model onto the input image and pruning of the parts, which are incompatible with the anthropomorphism. The orientation constraint is applied by creating synthetic views via regressing the input view to multiple oriented views. After applying the constraints, the 3D model is projected onto the initial and synthetic views, which further reduces the ambiguity. Finally, we borrow the direction of the unambiguous parts from the synthetic views to the initial one, which results in the 3D pose. Quantitative experiments are performed on the HumanEva-I dataset and qualitatively on unconstrained images from the Image Parse dataset. The results show the robustness of the proposed approach to accurately reconstruct the 3D pose form a single image.
Original languageEnglish
Title of host publicationProceedings of the IEEE International Conference on Computer Vision (ICCV2013)
EditorsKyros Kutulakos, Philip Torr, Steve Seitz, Yi Ma
Place of PublicationPiscataway, USA
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages1888-1895
Number of pages8
ISBN (Electronic)9781479928408
ISBN (Print)9781479930227
DOIs
Publication statusPublished - 2013
EventIEEE International Conference on Computer Vision (ICCV2013) - Sydney, Sydney, Australia
Duration: 1 Dec 20138 Dec 2013

Conference

ConferenceIEEE International Conference on Computer Vision (ICCV2013)
CountryAustralia
CitySydney
Period1/12/138/12/13

Fingerprint

Kinematics
Experiments

Cite this

Radwan, I., Dhall, A., & GOECKE, R. (2013). Monocular Image 3D Human Pose Estimation under Self-Occlusion. In K. Kutulakos, P. Torr, S. Seitz, & Y. Ma (Eds.), Proceedings of the IEEE International Conference on Computer Vision (ICCV2013) (pp. 1888-1895). Piscataway, USA: IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICCV.2013.237
Radwan, Ibrahim ; Dhall, Abhinav ; GOECKE, Roland. / Monocular Image 3D Human Pose Estimation under Self-Occlusion. Proceedings of the IEEE International Conference on Computer Vision (ICCV2013). editor / Kyros Kutulakos ; Philip Torr ; Steve Seitz ; Yi Ma. Piscataway, USA : IEEE, Institute of Electrical and Electronics Engineers, 2013. pp. 1888-1895
@inproceedings{87f6d38e3ceb4d429cae356afd50b1f8,
title = "Monocular Image 3D Human Pose Estimation under Self-Occlusion",
abstract = "In this paper, an automatic approach for 3D pose reconstruction from a single image is proposed. The presence of human body articulation, hallucinated parts and cluttered background leads to ambiguity during the pose inference, which makes the problem non-trivial. Researchers have explored various methods based on motion and shading in order to reduce the ambiguity and reconstruct the 3D pose. The key idea of our algorithm is to impose two types of constraints: kinematic constraints and orientation constraints. The kinematic constraints are imposed by projecting a 3D model onto the input image and pruning of the parts, which are incompatible with the anthropomorphism. The orientation constraint is applied by creating synthetic views via regressing the input view to multiple oriented views. After applying the constraints, the 3D model is projected onto the initial and synthetic views, which further reduces the ambiguity. Finally, we borrow the direction of the unambiguous parts from the synthetic views to the initial one, which results in the 3D pose. Quantitative experiments are performed on the HumanEva-I dataset and qualitatively on unconstrained images from the Image Parse dataset. The results show the robustness of the proposed approach to accurately reconstruct the 3D pose form a single image.",
keywords = "3D Pose estimation, Self-occlusion, Mixture of Pictorial Structures",
author = "Ibrahim Radwan and Abhinav Dhall and Roland GOECKE",
year = "2013",
doi = "10.1109/ICCV.2013.237",
language = "English",
isbn = "9781479930227",
pages = "1888--1895",
editor = "Kyros Kutulakos and Philip Torr and Steve Seitz and Yi Ma",
booktitle = "Proceedings of the IEEE International Conference on Computer Vision (ICCV2013)",
publisher = "IEEE, Institute of Electrical and Electronics Engineers",
address = "United States",

}

Radwan, I, Dhall, A & GOECKE, R 2013, Monocular Image 3D Human Pose Estimation under Self-Occlusion. in K Kutulakos, P Torr, S Seitz & Y Ma (eds), Proceedings of the IEEE International Conference on Computer Vision (ICCV2013). IEEE, Institute of Electrical and Electronics Engineers, Piscataway, USA, pp. 1888-1895, IEEE International Conference on Computer Vision (ICCV2013), Sydney, Australia, 1/12/13. https://doi.org/10.1109/ICCV.2013.237

Monocular Image 3D Human Pose Estimation under Self-Occlusion. / Radwan, Ibrahim; Dhall, Abhinav; GOECKE, Roland.

Proceedings of the IEEE International Conference on Computer Vision (ICCV2013). ed. / Kyros Kutulakos; Philip Torr; Steve Seitz; Yi Ma. Piscataway, USA : IEEE, Institute of Electrical and Electronics Engineers, 2013. p. 1888-1895.

Research output: A Conference proceeding or a Chapter in BookConference contribution

TY - GEN

T1 - Monocular Image 3D Human Pose Estimation under Self-Occlusion

AU - Radwan, Ibrahim

AU - Dhall, Abhinav

AU - GOECKE, Roland

PY - 2013

Y1 - 2013

N2 - In this paper, an automatic approach for 3D pose reconstruction from a single image is proposed. The presence of human body articulation, hallucinated parts and cluttered background leads to ambiguity during the pose inference, which makes the problem non-trivial. Researchers have explored various methods based on motion and shading in order to reduce the ambiguity and reconstruct the 3D pose. The key idea of our algorithm is to impose two types of constraints: kinematic constraints and orientation constraints. The kinematic constraints are imposed by projecting a 3D model onto the input image and pruning of the parts, which are incompatible with the anthropomorphism. The orientation constraint is applied by creating synthetic views via regressing the input view to multiple oriented views. After applying the constraints, the 3D model is projected onto the initial and synthetic views, which further reduces the ambiguity. Finally, we borrow the direction of the unambiguous parts from the synthetic views to the initial one, which results in the 3D pose. Quantitative experiments are performed on the HumanEva-I dataset and qualitatively on unconstrained images from the Image Parse dataset. The results show the robustness of the proposed approach to accurately reconstruct the 3D pose form a single image.

AB - In this paper, an automatic approach for 3D pose reconstruction from a single image is proposed. The presence of human body articulation, hallucinated parts and cluttered background leads to ambiguity during the pose inference, which makes the problem non-trivial. Researchers have explored various methods based on motion and shading in order to reduce the ambiguity and reconstruct the 3D pose. The key idea of our algorithm is to impose two types of constraints: kinematic constraints and orientation constraints. The kinematic constraints are imposed by projecting a 3D model onto the input image and pruning of the parts, which are incompatible with the anthropomorphism. The orientation constraint is applied by creating synthetic views via regressing the input view to multiple oriented views. After applying the constraints, the 3D model is projected onto the initial and synthetic views, which further reduces the ambiguity. Finally, we borrow the direction of the unambiguous parts from the synthetic views to the initial one, which results in the 3D pose. Quantitative experiments are performed on the HumanEva-I dataset and qualitatively on unconstrained images from the Image Parse dataset. The results show the robustness of the proposed approach to accurately reconstruct the 3D pose form a single image.

KW - 3D Pose estimation

KW - Self-occlusion

KW - Mixture of Pictorial Structures

U2 - 10.1109/ICCV.2013.237

DO - 10.1109/ICCV.2013.237

M3 - Conference contribution

SN - 9781479930227

SP - 1888

EP - 1895

BT - Proceedings of the IEEE International Conference on Computer Vision (ICCV2013)

A2 - Kutulakos, Kyros

A2 - Torr, Philip

A2 - Seitz, Steve

A2 - Ma, Yi

PB - IEEE, Institute of Electrical and Electronics Engineers

CY - Piscataway, USA

ER -

Radwan I, Dhall A, GOECKE R. Monocular Image 3D Human Pose Estimation under Self-Occlusion. In Kutulakos K, Torr P, Seitz S, Ma Y, editors, Proceedings of the IEEE International Conference on Computer Vision (ICCV2013). Piscataway, USA: IEEE, Institute of Electrical and Electronics Engineers. 2013. p. 1888-1895 https://doi.org/10.1109/ICCV.2013.237