# Gaussian Process Latent Variable Models For Human Pose Estimation

Carl Henrik Ek, University of Bristol
Philip H. S. Torr, University of Oxford
Neil D. Lawrence, University of Sheffield

in Machine Learning for Multimodal Interaction (MLMI 2007) 4892, pp 132-143

#### Abstract

We describe a method for recovering 3D human body pose from silhouettes. Our model is based on learning a latent space using the Gaussian Process Latent Variable Model (GP-LVM) [1] encapsulating both pose and silhouette features Our method is generative, this allows us to model the ambiguities of a silhouette representation in a principled way. We learn a dynamical model over the latent space which allows us to disambiguate between ambiguous silhouettes by temporal consistency. The model has only two free parameters and has several advantages over both regression approaches and other generative methods. In addition to the application shown in this paper the suggested model is easily extended to multiple observation spaces without constraints on type.

  @InProceedings{ek-pose07, title = {Gaussian Process Latent Variable Models For Human Pose Estimation}, author = {Carl Henrik Ek and Philip H. S. Torr and Neil D. Lawrence}, booktitle = {Machine Learning for Multimodal Interaction (MLMI 2007)}, pages = {132}, year = {2008}, editor = {Andrei Popescu-Belis and Steve Renals and Hervé Bourlard}, volume = {4892}, series = {LNCS}, address = {Brno, Czech Republic}, month = {00}, publisher = {Springer-Verlag}, edit = {https://github.com/lawrennd//publications/edit/gh-pages/_posts/2008-01-01-ek-pose07.md}, url = {http://inverseprobability.com/publications/ek-pose07.html}, abstract = {We describe a method for recovering 3D human body pose from silhouettes. Our model is based on learning a latent space using the Gaussian Process Latent Variable Model (GP-LVM) $1$ encapsulating both pose and silhouette features Our method is generative, this allows us to model the ambiguities of a silhouette representation in a principled way. We learn a dynamical model over the latent space which allows us to disambiguate between ambiguous silhouettes by temporal consistency. The model has only two free parameters and has several advantages over both regression approaches and other generative methods. In addition to the application shown in this paper the suggested model is easily extended to multiple observation spaces without constraints on type.}, key = {Ek:pose07}, doi = {10.1007/978-3-540-78155-4_12}, linkpdf = {ftp://ftp.dcs.shef.ac.uk/home/neil/mlmi.pdf}, linksoftware = {https://github.com/SheffieldML/SGPLVM/}, group = {gplvm,pose estimation} }
 %T Gaussian Process Latent Variable Models For Human Pose Estimation %A Carl Henrik Ek and Philip H. S. Torr and Neil D. Lawrence %B %C Machine Learning for Multimodal Interaction (MLMI 2007) %D %E Andrei Popescu-Belis and Steve Renals and Hervé Bourlard %F ek-pose07 %I Springer-Verlag %P 132--143 %R 10.1007/978-3-540-78155-4_12 %U http://inverseprobability.com/publications/ek-pose07.html %V 4892 %X We describe a method for recovering 3D human body pose from silhouettes. Our model is based on learning a latent space using the Gaussian Process Latent Variable Model (GP-LVM) $1$ encapsulating both pose and silhouette features Our method is generative, this allows us to model the ambiguities of a silhouette representation in a principled way. We learn a dynamical model over the latent space which allows us to disambiguate between ambiguous silhouettes by temporal consistency. The model has only two free parameters and has several advantages over both regression approaches and other generative methods. In addition to the application shown in this paper the suggested model is easily extended to multiple observation spaces without constraints on type. 
 TY - CPAPER TI - Gaussian Process Latent Variable Models For Human Pose Estimation AU - Carl Henrik Ek AU - Philip H. S. Torr AU - Neil D. Lawrence BT - Machine Learning for Multimodal Interaction (MLMI 2007) PY - 2008/01/01 DA - 2008/01/01 ED - Andrei Popescu-Belis ED - Steve Renals ED - Hervé Bourlard ID - ek-pose07 PB - Springer-Verlag SP - 132 EP - 143 DO - 10.1007/978-3-540-78155-4_12 L1 - ftp://ftp.dcs.shef.ac.uk/home/neil/mlmi.pdf UR - http://inverseprobability.com/publications/ek-pose07.html AB - We describe a method for recovering 3D human body pose from silhouettes. Our model is based on learning a latent space using the Gaussian Process Latent Variable Model (GP-LVM) $1$ encapsulating both pose and silhouette features Our method is generative, this allows us to model the ambiguities of a silhouette representation in a principled way. We learn a dynamical model over the latent space which allows us to disambiguate between ambiguous silhouettes by temporal consistency. The model has only two free parameters and has several advantages over both regression approaches and other generative methods. In addition to the application shown in this paper the suggested model is easily extended to multiple observation spaces without constraints on type. ER - 
 Ek, C.H., Torr, P.H.S. & Lawrence, N.D.. (2008). Gaussian Process Latent Variable Models For Human Pose Estimation. Machine Learning for Multimodal Interaction (MLMI 2007) 4892:132-143