Improving head and body pose estimation through semi-supervised manifold alignment

Handle URI:
http://hdl.handle.net/10754/556168
Title:
Improving head and body pose estimation through semi-supervised manifold alignment
Authors:
Heili, Alexandre; Varadarajan, Jagannadan; Ghanem, Bernard ( 0000-0002-5534-587X ) ; Ahuja, Narendra; Odobez, Jean-Marc
Abstract:
In this paper, we explore the use of a semi-supervised manifold alignment method for domain adaptation in the context of human body and head pose estimation in videos. We build upon an existing state-of-the-art system that leverages on external labelled datasets for the body and head features, and on the unlabelled test data with weak velocity labels to do a coupled estimation of the body and head pose. While this previous approach showed promising results, the learning of the underlying manifold structure of the features in the train and target data and the need to align them were not explored despite the fact that the pose features between two datasets may vary according to the scene, e.g. due to different camera point of view or perspective. In this paper, we propose to use a semi-supervised manifold alignment method to bring the train and target samples closer within the resulting embedded space. To this end, we consider an adaptation set from the target data and rely on (weak) labels, given for example by the velocity direction whenever they are reliable. These labels, along with the training labels are used to bias the manifold distance within each manifold and to establish correspondences for alignment.
KAUST Department:
Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
Citation:
Heili, Alexandre, Jagannadan Varadarajan, Bernard Ghanem, Narendra Ahuja, and Jean-Marc Odobez. "Improving Head and Body Pose Estimation through Semi-supervised Manifold Alignment." In International Conference on Image Processing, no. EPFL-CONF-200303. 2014.
Publisher:
IEEE
Journal:
Image Processing (ICIP), 2014 IEEE International Conference on
Conference/Event name:
2014 IEEE International Conference on Image Processing, ICIP 2014
Issue Date:
27-Oct-2014
DOI:
10.1109/ICIP.2014.7025383
Type:
Conference Paper
Sponsors:
IEEE TERMS Couplings Estimation Manifolds Surveillance Training Vectors Videos
Additional Links:
http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=7025383; https://dl.dropboxusercontent.com/u/18955644/website_files/Heili_ICIP_2014.pdf
Appears in Collections:
Conference Papers; Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division

Full metadata record

DC FieldValue Language
dc.contributor.authorHeili, Alexandreen
dc.contributor.authorVaradarajan, Jagannadanen
dc.contributor.authorGhanem, Bernarden
dc.contributor.authorAhuja, Narendraen
dc.contributor.authorOdobez, Jean-Marcen
dc.date.accessioned2015-06-02T14:24:24Zen
dc.date.available2015-06-02T14:24:24Zen
dc.date.issued2014-10-27en
dc.identifier.citationHeili, Alexandre, Jagannadan Varadarajan, Bernard Ghanem, Narendra Ahuja, and Jean-Marc Odobez. "Improving Head and Body Pose Estimation through Semi-supervised Manifold Alignment." In International Conference on Image Processing, no. EPFL-CONF-200303. 2014.en
dc.identifier.doi10.1109/ICIP.2014.7025383en
dc.identifier.urihttp://hdl.handle.net/10754/556168en
dc.description.abstractIn this paper, we explore the use of a semi-supervised manifold alignment method for domain adaptation in the context of human body and head pose estimation in videos. We build upon an existing state-of-the-art system that leverages on external labelled datasets for the body and head features, and on the unlabelled test data with weak velocity labels to do a coupled estimation of the body and head pose. While this previous approach showed promising results, the learning of the underlying manifold structure of the features in the train and target data and the need to align them were not explored despite the fact that the pose features between two datasets may vary according to the scene, e.g. due to different camera point of view or perspective. In this paper, we propose to use a semi-supervised manifold alignment method to bring the train and target samples closer within the resulting embedded space. To this end, we consider an adaptation set from the target data and rely on (weak) labels, given for example by the velocity direction whenever they are reliable. These labels, along with the training labels are used to bias the manifold distance within each manifold and to establish correspondences for alignment.en
dc.description.sponsorshipIEEE TERMS Couplings Estimation Manifolds Surveillance Training Vectors Videosen
dc.publisherIEEEen
dc.relation.urlhttp://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=7025383en
dc.relation.urlhttps://dl.dropboxusercontent.com/u/18955644/website_files/Heili_ICIP_2014.pdfen
dc.rights(c) 2014 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/ republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of any copyrighted components of this work in other works.en
dc.subjectdomain adaptationen
dc.subjecthead and body poseen
dc.subjectweak labelsen
dc.subjectsurveillanceen
dc.subjectsemi-superviseden
dc.subjectmanifolden
dc.titleImproving head and body pose estimation through semi-supervised manifold alignmenten
dc.typeConference Paperen
dc.contributor.departmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Divisionen
dc.identifier.journalImage Processing (ICIP), 2014 IEEE International Conference onen
dc.conference.date27 October 2014 through 30 October 2014en
dc.conference.name2014 IEEE International Conference on Image Processing, ICIP 2014en
dc.eprint.versionPost-printen
dc.contributor.institutionUniversity of California San Diego, USAen
kaust.authorGhanem, Bernarden
All Items in KAUST are protected by copyright, with all rights reserved, unless otherwise indicated.