Open Access

3D Shape-Encoded Particle Filter for Object Tracking and Its Application to Human Body Tracking

EURASIP Journal on Image and Video Processing20072008:596989

https://doi.org/10.1155/2008/596989

Received: 1 February 2007

Accepted: 25 November 2007

Published: 25 November 2007

Abstract

We present a nonlinear state estimation approach using particle filters, for tracking objects whose approximate 3D shapes are known. The unnormalized conditional density for the solution to the nonlinear filtering problem leads to the Zakai equation, and is realized by the weights of the particles. The weight of a particle represents its geometric and temporal fit, which is computed bottom-up from the raw image using a shape-encoded filter. The main contribution of the paper is the design of smoothing filters for feature extraction combined with the adoption of unnormalized conditional density weights. The "shape filter" has the overall form of the predicted 2D projection of the 3D model, while the cross-section of the filter is designed to collect the gradient responses along the shape. The 3D-model-based representation is designed to emphasize the changes in 2D object shape due to motion, while de-emphasizing the variations due to lighting and other imaging conditions. We have found that the set of sparse measurements using a relatively small number of particles is able to approximate the high-dimensional state distribution very effectively. As a measures to stabilize the tracking, the amount of random diffusion is effectively adjusted using a Kalman updating of the covariance matrix. For a complex problem of human body tracking, we have successfully employed constraints derived from joint angles and walking motion.

Publisher note

To access the full article, please see PDF.

Authors’ Affiliations

(1)
VideoMining Corporation
(2)
Department of Electrical and Computer Engineering, University of Maryland

Copyright

© H. Moon and R. Chellappa. 2008

This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.