Tracking Using Continuous Shape Model Learning in the Presence of Occlusion
© M. Asadi and C. S. Regazzoni. 2008
Received: 1 April 2007
Accepted: 17 January 2008
Published: 12 February 2008
This paper presents a Bayesian framework for a new model-based learning method, which is able to track nonrigid objects in the presence of occlusions, based on a dynamic shape description in terms of a set of corners. Tracking is done by estimating the new position of the target in a multimodal voting space. However, occlusion events and clutter may affect the model learning, leading to a distraction in the estimation of the new position of the target as well as incorrect updating of the shape model. This method takes advantage of automatic decisions regarding how to learn the model in different environments, by estimating the possible presence of distracters and regulating corner updating on the basis of these estimations. Moreover, by introducing the corner feature vector classification, the method is able to continue learning the model dynamically, even in such situations. Experimental results show a successful tracking along with a more precise estimation of shape and motion during occlusion events.
To access the full article, please see PDF.
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.