Visual contour tracking based on sequential importance sampling/resampling algorithm

The condensation algorithm can deal with non-Gaussian, nonlinear visual contour tracking in a unified way. Despite its simple implementation and generality, it has two main limitations. The first limitation is that in sampling stage the algorithm does not take advantage of the new measurements. As a result of the inefficient sampling strategy, the algorithm needs a large number of samples to represent the posterior distribution of state. The next is in the selection step, resampling may introduce the problem of sample impoverishment. To address these two problems, we present an improved visual tracker based on an importance sampling/resampling algorithm. Gaussian density of each sample is adopted as the sub-optimal importance proposal distribution, which can steer the samples towards the high likelihood by considering the latest observations. We also adopt a criterion of effective sample size to determine whether the resampling is necessary or not. Experiments with real image sequences show that the performance of new algorithm improves considerably for tracking in visual clutter.

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

[2]  Simon J. Godsill,et al.  On sequential simulation-based methods for Bayesian filtering , 1998 .

[3]  Natan Peterfreund,et al.  Robust Tracking of Position and Velocity With Kalman Snakes , 1999, IEEE Trans. Pattern Anal. Mach. Intell..

[4]  Nando de Freitas,et al.  Sequential Monte Carlo Methods in Practice , 2001, Statistics for Engineering and Information Science.

[5]  Jun S. Liu,et al.  Sequential Monte Carlo methods for dynamic systems , 1997 .

[6]  Richard Szeliski,et al.  Tracking with Kalman snakes , 1993 .

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

[8]  Michael Isard,et al.  Learning to Track the Visual Motion of Contours , 1995, Artif. Intell..

[9]  Arnaud Doucet,et al.  Sequential Monte Carlo Methods to Train Neural Network Models , 2000, Neural Computation.

[10]  Andrew Blake,et al.  A Probabilistic Exclusion Principle for Tracking Multiple Objects , 2004, International Journal of Computer Vision.

[11]  Michael Isard,et al.  CONDENSATION—Conditional Density Propagation for Visual Tracking , 1998, International Journal of Computer Vision.

[12]  Y. Bar-Shalom.  Tracking and data association , 1988 .

[13]  Michael Isard,et al.  Partitioned Sampling, Articulated Objects, and Interface-Quality Hand Tracking , 2000, ECCV.

[14]  Dimitris N. Metaxas.  Shape and Nonrigid Motion Estimation , 1997 .

[15]  Michael Isard,et al.  ICONDENSATION: Unifying Low-Level and High-Level Tracking in a Stochastic Framework , 1998, ECCV.

[16]  M. Pitt,et al.  Filtering via Simulation: Auxiliary Particle Filters , 1999 .

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