Face Tracking Algorithm Based on Sequential Monte Carlo Filter

Abstract:

Incorporating color distribution and spatial layout, this paper proposes a sequential Monte Carlo filter posterior tracking algorithm using color and spatial information in HSV color space. The target model is defined by the spatial color information of the tracking face region. By computing the characteristic distance between sample and target, different weights associated with every sample and the posterior of state vector can be computed. The samples distribution trends to the state distribution, whose validity is guaranteed by the strong law of large numbers. The tracking results using weighted samples are given in simulation. Experimental results show the probabilistic approach is simple and computationally efficient. In addition, this algorithm based on the sequential Monte Carlo filter could predict the location of face and track its trajectory satisfactorily in various complex conditions.

• Periodical: Advanced Materials Research (Volumes 430-432)
• Pages: 1777-1781
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.430-432.1777
• Online since: January 2012
• Authors: Yun Ming Du, Bing Bing Yan, Yong Cheng Jiang
• Keywords: Bayesian Estimation, Posterior Tracking, Sequential Monte Carlo Filter, Weighted Samples

References

[1] S.T. Huang, N. Sebe, M.S. Lew et al.: Computer Vision in Human-Computer Interaction(Springer Verlag, New York 2005).

[2] A.J. Haug: MTR 05W0000004(2005).

[3] M. Beal, Z. Ghahramani, and C. Rasmussen: Advances in Neural Information Processing Systems Vol. 14 (2002), pp.577-584.

[4] G. Mao, S. Drake and B.D.O. Anderson: Information Decision and Control Vol. 1 (2007), pp.224-229.

[5] A. Doucet, J.F.G. Freitas and N.J. Gordon: Sequential Monte Carlo Methods in Practice(Springer Verlag, New York 2001).

[6] P. Brasnett, L. Mihaylova, D. Bull et al.: Image and Vision Computing Archive Vol. 25 (2007), pp.1217-1227.