Gibbs likelihoods for Bayesian tracking

Bayesian methods for visual tracking model the likelihood of image measurements conditioned on a tracking hypothesis. Image measurements may, for example, correspond to various filter responses at multiple scales and orientations. Most tracking approaches exploit ad hoc likelihood models while those that exploit more rigorous, learned, models often make unrealistic assumptions about the underlying probabilistic model. Such assumptions cause problems for Bayesian inference when an unsound likelihood is combined with an a priori probability distribution. Errors in modeling the likelihood can lead to brittle tracking results, particularly when using non-parametric inference techniques such as particle filtering. We show how assumptions of conditional independence of filter responses are violated in common tracking scenarios, lead to incorrect likelihood models, and cause problems for Bayesian inference. We address the problem of modeling more principled likelihoods using Gibbs learning. The learned models are compared with naive Bayes methods which assume conditional independence of the filter responses. We show how these Gibbs models can be used as an effective image likelihood, and demonstrate them in the context of particle filter-based human tracking.

[1]  Jia Jie Bayesian denoising of visual images in the wavelet domain , 2003 .

[2]  D. Ruderman The statistics of natural images , 1994 .

[3]  J. Friedman,et al.  PROJECTION PURSUIT DENSITY ESTIMATION , 1984 .

[4]  Song-Chun Zhu,et al.  Prior Learning and Gibbs Reaction-Diffusion , 1997, IEEE Trans. Pattern Anal. Mach. Intell..

[5]  Michael Isard,et al.  Contour Tracking by Stochastic Propagation of Conditional Density , 1996, ECCV.

[6]  David A. Forsyth,et al.  How Does CONDENSATION Behave with a Finite Number of Samples? , 2000, ECCV.

[7]  Michael J. Black,et al.  Learning the Statistics of People in Images and Video , 2003, International Journal of Computer Vision.

[8]  Eero P. Simoncelli,et al.  Image Denoising using Gaussian Scale Mixtures in the Wavelet Domain , 2002 .

[9]  Harry Shum,et al.  An efficient approach to learning inhomogeneous Gibbs model , 2003, 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003. Proceedings..

[10]  Michael Isard,et al.  BraMBLe: a Bayesian multiple-blob tracker , 2001, Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001.

[11]  N. Gordon,et al.  Novel approach to nonlinear/non-Gaussian Bayesian state estimation , 1993 .

[12]  Sidharth Bhatia,et al.  Tracking loose-limbed people , 2004, Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004..

[13]  Song-Chun Zhu,et al.  FRAME: filters, random fields, and minimax entropy towards a unified theory for texture modeling , 1996, Proceedings CVPR IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[14]  Andrew Blake,et al.  Statistical Foreground Modelling for Object Localisation , 2000, ECCV.

[15]  Andrew Blake,et al.  A Probabilistic Background Model for Tracking , 2000, ECCV.

[16]  Michael Isard,et al.  Object localization by Bayesian correlation , 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision.

[17]  Geoffrey E. Hinton,et al.  Efficient Parametric Projection Pursuit Density Estimation , 2003, UAI.