Adaptive occlusion state estimation for human pose tracking under self-occlusions

Nam Gyu Cho, Alan L. Yuille, Seong Whan Lee

Research output: Contribution to journalArticle

21 Citations (Scopus)

Abstract

Tracking human poses in video can be considered as the process of inferring the positions of the body joints. Among various obstacles to this task, one of the most challenging is to deal with 'self-occlusion?, where one body part occludes another one. In order to tackle this problem, a model must represent the self-occlusion between different body parts which leads to complex inference problems. In this paper, we propose a method that estimates occlusion states adaptively. A Markov random field is used to represent the occlusion relationship between human body parts in terms an occlusion state variable, which represents the depth order. To ensure efficient computation, inference is divided into two steps: a body pose inference step and an occlusion state inference step. We test our method using video sequences from the HumanEva dataset. We label the data to quantify how the relative depth ordering of parts, and hence the self-occlusion, changes during the video sequence. Then we demonstrate that our method can successfully track human poses even when there are frequent occlusion changes. We compare our approach to alternative methods including the state of the art approach which use multiple cameras.

Original languageEnglish
Pages (from-to)649-661
Number of pages13
JournalPattern Recognition
Volume46
Issue number3
DOIs
Publication statusPublished - 2013 Mar 1

Fingerprint

State estimation
Labels
Cameras

Keywords

  • 3D human pose tracking
  • Computer vision
  • Self-occlusion

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Signal Processing

Cite this

Adaptive occlusion state estimation for human pose tracking under self-occlusions. / Cho, Nam Gyu; Yuille, Alan L.; Lee, Seong Whan.

In: Pattern Recognition, Vol. 46, No. 3, 01.03.2013, p. 649-661.

Research output: Contribution to journalArticle

@article{508811fc32b84e3aa770ac99fd74763f,
title = "Adaptive occlusion state estimation for human pose tracking under self-occlusions",
abstract = "Tracking human poses in video can be considered as the process of inferring the positions of the body joints. Among various obstacles to this task, one of the most challenging is to deal with 'self-occlusion?, where one body part occludes another one. In order to tackle this problem, a model must represent the self-occlusion between different body parts which leads to complex inference problems. In this paper, we propose a method that estimates occlusion states adaptively. A Markov random field is used to represent the occlusion relationship between human body parts in terms an occlusion state variable, which represents the depth order. To ensure efficient computation, inference is divided into two steps: a body pose inference step and an occlusion state inference step. We test our method using video sequences from the HumanEva dataset. We label the data to quantify how the relative depth ordering of parts, and hence the self-occlusion, changes during the video sequence. Then we demonstrate that our method can successfully track human poses even when there are frequent occlusion changes. We compare our approach to alternative methods including the state of the art approach which use multiple cameras.",
keywords = "3D human pose tracking, Computer vision, Self-occlusion",
author = "Cho, {Nam Gyu} and Yuille, {Alan L.} and Lee, {Seong Whan}",
year = "2013",
month = "3",
day = "1",
doi = "10.1016/j.patcog.2012.09.006",
language = "English",
volume = "46",
pages = "649--661",
journal = "Pattern Recognition",
issn = "0031-3203",
publisher = "Elsevier Limited",
number = "3",

}

TY - JOUR

T1 - Adaptive occlusion state estimation for human pose tracking under self-occlusions

AU - Cho, Nam Gyu

AU - Yuille, Alan L.

AU - Lee, Seong Whan

PY - 2013/3/1

Y1 - 2013/3/1

N2 - Tracking human poses in video can be considered as the process of inferring the positions of the body joints. Among various obstacles to this task, one of the most challenging is to deal with 'self-occlusion?, where one body part occludes another one. In order to tackle this problem, a model must represent the self-occlusion between different body parts which leads to complex inference problems. In this paper, we propose a method that estimates occlusion states adaptively. A Markov random field is used to represent the occlusion relationship between human body parts in terms an occlusion state variable, which represents the depth order. To ensure efficient computation, inference is divided into two steps: a body pose inference step and an occlusion state inference step. We test our method using video sequences from the HumanEva dataset. We label the data to quantify how the relative depth ordering of parts, and hence the self-occlusion, changes during the video sequence. Then we demonstrate that our method can successfully track human poses even when there are frequent occlusion changes. We compare our approach to alternative methods including the state of the art approach which use multiple cameras.

AB - Tracking human poses in video can be considered as the process of inferring the positions of the body joints. Among various obstacles to this task, one of the most challenging is to deal with 'self-occlusion?, where one body part occludes another one. In order to tackle this problem, a model must represent the self-occlusion between different body parts which leads to complex inference problems. In this paper, we propose a method that estimates occlusion states adaptively. A Markov random field is used to represent the occlusion relationship between human body parts in terms an occlusion state variable, which represents the depth order. To ensure efficient computation, inference is divided into two steps: a body pose inference step and an occlusion state inference step. We test our method using video sequences from the HumanEva dataset. We label the data to quantify how the relative depth ordering of parts, and hence the self-occlusion, changes during the video sequence. Then we demonstrate that our method can successfully track human poses even when there are frequent occlusion changes. We compare our approach to alternative methods including the state of the art approach which use multiple cameras.

KW - 3D human pose tracking

KW - Computer vision

KW - Self-occlusion

UR - http://www.scopus.com/inward/record.url?scp=84870249775&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84870249775&partnerID=8YFLogxK

U2 - 10.1016/j.patcog.2012.09.006

DO - 10.1016/j.patcog.2012.09.006

M3 - Article

AN - SCOPUS:84870249775

VL - 46

SP - 649

EP - 661

JO - Pattern Recognition

JF - Pattern Recognition

SN - 0031-3203

IS - 3

ER -