Developing a Bayesian framework for human behavior tracking

抄録

In this paper, we study the problem of creating an inference mechanism to recognize and respond to human behavior. We provide probabilistic methods to build a new Bayesian framework to deal with human tracking problem. Specifically, we present a set of efficient algorithms that encompass the learning solutions for practical applications which cope with unreliable and noisy measurements. Unlike almost all of related works, we propose an efficient algorithm for sensing systems that presents an alternative to sensors that are sometimes perceived as invasive, where notably we do not use vision-based learning. Preliminary results show that the proposed system can be deployed in different environments and significantly outperforms existing methods in a very reliable manner.