A robust statistical approach to example-based pose estimation






Fabio Cuzzolin,
To submit to the International Journal of Computer Vision, 2011
Abstract

Pose estimation concerns the reconstruction of the pose of a moving object from a sequence of images shot during its motion. When having no models or any a-priori information about the nature of the body, inference on the object’s pose requires building a map between image measurements (features) and poses. In this paper we present a method for pose estimation based on the evidential reasoning, in which an “evidential model” of the object is learned from a set of examples in a training stage. We use hidden Markov models to find a multi-modal Gaussian representation of the involved feature spaces and learn feature-pose maps from the training data. In the estimation stage all the features coming from one or more views are expressed as belief functions on the respective feature spaces, projected, and combined in an approximate parameter space, improving estimation robustness and precision. Experimental results concerning the two-view human tracking problem are shown.
sp 
Download
PDF  Zipped Postscript

BibTeX entry

@article{cuzzolin11ijcv-pose, 
  AUTHOR = "Fabio Cuzzolin", 
  TITLE = "A robust statistical approach to example-based pose estimation ",
  JOURNAL = "to submit to the International Journal of Computer Vision", 
  YEAR = "2011" 
}