A Bayesian approach to groups of people tracking

Abstract

We proposed a novel approach to track groups of people observed by a static camera. The approach relies on a unified Bayes' theorem based framework to model both scene background and the color distribution of targets. By sharing the framework, it is more simple and low cost to construct a practical surveillance system. Additionally, the framework has the advantages of insensitiveness to initial observations and the capability of adaptive selection of modal number compared with GMM framework. By modeling the color distribution of people, the approach can keep tracking not only single person targets but also those occluded by or interacting with other people in the scene. The experiments using outdoor videos prove that the approach is effective.