Domain adaptation for upper body pose tracking in signed TV broadcasts
The objective of this work is to estimate upper body pose for signers in TV broadcasts. Given suitable training data, the pose is estimated using a random forest body joint detector. However, obtaining such training data can be costly. The novelty of this paper is a method of transfer learning which...
Main Authors: | , , , , |
---|---|
Format: | Conference item |
Published: |
British Machine Vision Association
2013
|
_version_ | 1797080435886391296 |
---|---|
author | Charles, J Pfister, T Magee, D Hogg, D Zisserman, A |
author_facet | Charles, J Pfister, T Magee, D Hogg, D Zisserman, A |
author_sort | Charles, J |
collection | OXFORD |
description | The objective of this work is to estimate upper body pose for signers in TV broadcasts. Given suitable training data, the pose is estimated using a random forest body joint detector. However, obtaining such training data can be costly. The novelty of this paper is a method of transfer learning which is able to harness existing training data and use it for new domains. Our contributions are: (i) a method for adapting existing training data to generate new training data by synthesis for signers with different appearances, and (ii) a method for personalising training data. As a case study we show how the appearance of the arms for different clothing, specifically short and long sleeved clothes, can be modelled to obtain person-specific trackers. We demonstrate that the transfer learning and person specific trackers significantly improve pose estimation performance. |
first_indexed | 2024-03-07T01:00:00Z |
format | Conference item |
id | oxford-uuid:8962dec7-c750-4953-b9d0-b9d545a33859 |
institution | University of Oxford |
last_indexed | 2024-03-07T01:00:00Z |
publishDate | 2013 |
publisher | British Machine Vision Association |
record_format | dspace |
spelling | oxford-uuid:8962dec7-c750-4953-b9d0-b9d545a338592022-03-26T22:24:13ZDomain adaptation for upper body pose tracking in signed TV broadcastsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:8962dec7-c750-4953-b9d0-b9d545a33859Symplectic Elements at OxfordBritish Machine Vision Association2013Charles, JPfister, TMagee, DHogg, DZisserman, AThe objective of this work is to estimate upper body pose for signers in TV broadcasts. Given suitable training data, the pose is estimated using a random forest body joint detector. However, obtaining such training data can be costly. The novelty of this paper is a method of transfer learning which is able to harness existing training data and use it for new domains. Our contributions are: (i) a method for adapting existing training data to generate new training data by synthesis for signers with different appearances, and (ii) a method for personalising training data. As a case study we show how the appearance of the arms for different clothing, specifically short and long sleeved clothes, can be modelled to obtain person-specific trackers. We demonstrate that the transfer learning and person specific trackers significantly improve pose estimation performance. |
spellingShingle | Charles, J Pfister, T Magee, D Hogg, D Zisserman, A Domain adaptation for upper body pose tracking in signed TV broadcasts |
title | Domain adaptation for upper body pose tracking in signed TV broadcasts |
title_full | Domain adaptation for upper body pose tracking in signed TV broadcasts |
title_fullStr | Domain adaptation for upper body pose tracking in signed TV broadcasts |
title_full_unstemmed | Domain adaptation for upper body pose tracking in signed TV broadcasts |
title_short | Domain adaptation for upper body pose tracking in signed TV broadcasts |
title_sort | domain adaptation for upper body pose tracking in signed tv broadcasts |
work_keys_str_mv | AT charlesj domainadaptationforupperbodyposetrackinginsignedtvbroadcasts AT pfistert domainadaptationforupperbodyposetrackinginsignedtvbroadcasts AT mageed domainadaptationforupperbodyposetrackinginsignedtvbroadcasts AT hoggd domainadaptationforupperbodyposetrackinginsignedtvbroadcasts AT zissermana domainadaptationforupperbodyposetrackinginsignedtvbroadcasts |