Pushpa Publishing House

Journal Menu

Content

Volume 17 (2018)

Volume 16 (2016)

Volume 15 (2015)

Volume 14 (2015)

Volume 13 (2014)

Volume 12 (2014)

Volume 11 (2013)

Volume 10 (2013)

Volume 9 (2012)

Volume 8 (2012)

Volume 7 (2011)

Volume 6 (2011)

Volume 5 (2010)

	Volume 5, Issue 2 Pg 131 - 263 (November 2010)
	Volume 5, Issue 1 Pg 1 - 129 (August 2010)

Volume 4 (2010)

Volume 3 (2009)

Volume 2 (2008)

Volume 1 (2007)

Advances in Computer Science and Engineering

Advances in Computer Science and Engineering
Volume 5, Issue 1, Pages 25 - 35 (August 2010)

MODEL UPDATE FOR OBJECT TRACKING WITH PARTICLE FILTER
Wang Shu-Peng

Abstract:
A novel model update method in the framework of particle filter is proposed in this paper to deal with the variation of target appearance during tracking. Particle filters are sequential Monte Carlo methods which approximate the probability densities with a weighted sample set. The proposed algorithm efficiently utilizes the sample set and target appearance information for updating the target model. This algorithm involves the estimation for tracking reliability. To avoid model “drift”, the target model is updated only during the reliable tracking and the update speed is constrained by the tracking reliability. Tracking reliability is determined from the shape of the probability density and similarity between observation image and target model. Tracking is considered as reliability when the probability density tends to have single major mode and the similarity between the observation image and the target model is greater than a predefined threshold. The modes of the probability density are estimated using the mean shift methods. The proposed algorithm avoids the model drift caused by the naive algorithm. Experimental results are shown to validate the effectiveness of the proposed algorithm.

Keywords and phrases: object tracking, particle filter, model update, mean shift, appearance model, clustering.

Number of Downloads: 451 | Number of Views: 833