Variational Inference for Visual Tracking

[edit]

Jaco Vermaak
Neil D. Lawrence, University of Sheffield
Patrick Pérez

in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition I, pp 773-780

Related Material

Abstract

The likelihood models used in probabilistic visual tracking applications are often complex non-linear and/or non-Gaussian functions, leading to analytically intractable inference. Solutions then require numerical approximation techniques, of which the particle filter is a popular choice. Particle filters, however, degrade in performance as the dimensionality of the state space increases and the support of the likelihood decreases. As an alternative to particle filters this paper introduces a variational approximation to the tracking recursion. The variational inference is intractable in itself, and is combined with an efficient importance sampling procedure to obtain the required estimates. The algorithm is shown to compare favourably with particle filtering techniques on a synthetic example and two real tracking problems. The first involves the tracking of a designated object in a video sequence based on its colour properties, whereas the second involves contour extraction in a single image.


@InProceedings{vermaak-variational03,
  title = 	 {Variational Inference for Visual Tracking},
  author = 	 {Jaco Vermaak and Neil D. Lawrence and Patrick Pérez},
  booktitle = 	 {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition},
  pages = 	 {773},
  year = 	 {2003},
  volume = 	 {I},
  month = 	 {00},
  publisher = 	 {IEEE Computer Society Press},
  edit = 	 {https://github.com/lawrennd//publications/edit/gh-pages/_posts/2003-01-01-vermaak-variational03.md},
  url =  	 {http://inverseprobability.com/publications/vermaak-variational03.html},
  abstract = 	 {The likelihood models used in probabilistic visual tracking applications are often complex non-linear and/or non-Gaussian functions, leading to analytically intractable inference. Solutions then require numerical approximation techniques, of which the particle filter is a popular choice. Particle filters, however, degrade in performance as the dimensionality of the state space increases and the support of the likelihood decreases. As an alternative to particle filters this paper introduces a variational approximation to the tracking recursion. The variational inference is intractable in itself, and is combined with an efficient importance sampling procedure to obtain the required estimates. The algorithm is shown to compare favourably with particle filtering techniques on a synthetic example and two real tracking problems. The first involves the tracking of a designated object in a video sequence based on its colour properties, whereas the second involves contour extraction in a single image.},
  key = 	 {Vermaak:variational03},
  linkpdf = 	 {ftp://ftp.dcs.shef.ac.uk/home/neil/variationalTracking.pdf},
  group = 	 {shefml}
 

}
%T Variational Inference for Visual Tracking
%A Jaco Vermaak and Neil D. Lawrence and Patrick Pérez
%B 
%C Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
%D 
%F vermaak-variational03
%I IEEE Computer Society Press	
%P 773--780
%R 
%U http://inverseprobability.com/publications/vermaak-variational03.html
%V I
%X The likelihood models used in probabilistic visual tracking applications are often complex non-linear and/or non-Gaussian functions, leading to analytically intractable inference. Solutions then require numerical approximation techniques, of which the particle filter is a popular choice. Particle filters, however, degrade in performance as the dimensionality of the state space increases and the support of the likelihood decreases. As an alternative to particle filters this paper introduces a variational approximation to the tracking recursion. The variational inference is intractable in itself, and is combined with an efficient importance sampling procedure to obtain the required estimates. The algorithm is shown to compare favourably with particle filtering techniques on a synthetic example and two real tracking problems. The first involves the tracking of a designated object in a video sequence based on its colour properties, whereas the second involves contour extraction in a single image.
TY  - CPAPER
TI  - Variational Inference for Visual Tracking
AU  - Jaco Vermaak
AU  - Neil D. Lawrence
AU  - Patrick Pérez
BT  - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
PY  - 2003/01/01
DA  - 2003/01/01	
ID  - vermaak-variational03
PB  - IEEE Computer Society Press	
SP  - 773
EP  - 780
L1  - ftp://ftp.dcs.shef.ac.uk/home/neil/variationalTracking.pdf
UR  - http://inverseprobability.com/publications/vermaak-variational03.html
AB  - The likelihood models used in probabilistic visual tracking applications are often complex non-linear and/or non-Gaussian functions, leading to analytically intractable inference. Solutions then require numerical approximation techniques, of which the particle filter is a popular choice. Particle filters, however, degrade in performance as the dimensionality of the state space increases and the support of the likelihood decreases. As an alternative to particle filters this paper introduces a variational approximation to the tracking recursion. The variational inference is intractable in itself, and is combined with an efficient importance sampling procedure to obtain the required estimates. The algorithm is shown to compare favourably with particle filtering techniques on a synthetic example and two real tracking problems. The first involves the tracking of a designated object in a video sequence based on its colour properties, whereas the second involves contour extraction in a single image.
ER  -

Vermaak, J., Lawrence, N.D. & Pérez, P.. (2003). Variational Inference for Visual Tracking. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition I:773-780