Backing Off: Hierarchical Decomposition of Activity for 3D Novel Pose Recovery

John Darby, Baihua Li, Nicholas Costen, David J. Fleet, Neil D. Lawrence
, 2009.

Abstract

For model-based 3D human pose estimation, even simple models of the human body lead to high-dimensional state spaces. Where the class of activity is known a priori, lowdimensional activity models learned from training data make possible a thorough and efficient search for the best pose. Conversely, searching for solutions in the full state space places no restriction on the class of motion to be recovered, but is both difficult and expensive. This paper explores a potential middle ground between these approaches, using the hierarchical Gaussian process latent variable model to learn activity at different hierarchical scales within the human skeleton. We show that by training on full-body activity data then descending through the hierarchy in stages and exploring subtrees independently of one another, novel poses may be recovered. Experimental results on motion capture data and monocular video sequences demonstrate the utility of the approach, and comparisons are drawn with existing low-dimensional activity models

Cite this Paper


BibTeX
@InProceedings{pmlr-v-darby-backing09, title = {Backing Off: Hierarchical Decomposition of Activity for 3D Novel Pose Recovery}, author = {John Darby and Baihua Li and Nicholas Costen and David J. Fleet and Neil D. Lawrence}, year = {}, editor = {}, url = {http://inverseprobability.com/publications/darby-backing09.html}, abstract = {For model-based 3D human pose estimation, even simple models of the human body lead to high-dimensional state spaces. Where the class of activity is known a priori, lowdimensional activity models learned from training data make possible a thorough and efficient search for the best pose. Conversely, searching for solutions in the full state space places no restriction on the class of motion to be recovered, but is both difficult and expensive. This paper explores a potential middle ground between these approaches, using the hierarchical Gaussian process latent variable model to learn activity at different hierarchical scales within the human skeleton. We show that by training on full-body activity data then descending through the hierarchy in stages and exploring subtrees independently of one another, novel poses may be recovered. Experimental results on motion capture data and monocular video sequences demonstrate the utility of the approach, and comparisons are drawn with existing low-dimensional activity models} }
Endnote
%0 Conference Paper %T Backing Off: Hierarchical Decomposition of Activity for 3D Novel Pose Recovery %A John Darby %A Baihua Li %A Nicholas Costen %A David J. Fleet %A Neil D. Lawrence %B %C Proceedings of Machine Learning Research %D %E %F pmlr-v-darby-backing09 %I PMLR %J Proceedings of Machine Learning Research %P -- %U http://inverseprobability.com %V %W PMLR %X For model-based 3D human pose estimation, even simple models of the human body lead to high-dimensional state spaces. Where the class of activity is known a priori, lowdimensional activity models learned from training data make possible a thorough and efficient search for the best pose. Conversely, searching for solutions in the full state space places no restriction on the class of motion to be recovered, but is both difficult and expensive. This paper explores a potential middle ground between these approaches, using the hierarchical Gaussian process latent variable model to learn activity at different hierarchical scales within the human skeleton. We show that by training on full-body activity data then descending through the hierarchy in stages and exploring subtrees independently of one another, novel poses may be recovered. Experimental results on motion capture data and monocular video sequences demonstrate the utility of the approach, and comparisons are drawn with existing low-dimensional activity models
RIS
TY - CPAPER TI - Backing Off: Hierarchical Decomposition of Activity for 3D Novel Pose Recovery AU - John Darby AU - Baihua Li AU - Nicholas Costen AU - David J. Fleet AU - Neil D. Lawrence BT - PY - DA - ED - ID - pmlr-v-darby-backing09 PB - PMLR SP - DP - PMLR EP - L1 - UR - http://inverseprobability.com/publications/darby-backing09.html AB - For model-based 3D human pose estimation, even simple models of the human body lead to high-dimensional state spaces. Where the class of activity is known a priori, lowdimensional activity models learned from training data make possible a thorough and efficient search for the best pose. Conversely, searching for solutions in the full state space places no restriction on the class of motion to be recovered, but is both difficult and expensive. This paper explores a potential middle ground between these approaches, using the hierarchical Gaussian process latent variable model to learn activity at different hierarchical scales within the human skeleton. We show that by training on full-body activity data then descending through the hierarchy in stages and exploring subtrees independently of one another, novel poses may be recovered. Experimental results on motion capture data and monocular video sequences demonstrate the utility of the approach, and comparisons are drawn with existing low-dimensional activity models ER -
APA
Darby, J., Li, B., Costen, N., Fleet, D.J. & Lawrence, N.D.. (). Backing Off: Hierarchical Decomposition of Activity for 3D Novel Pose Recovery. , in PMLR :-

Related Material