Tracking Multiple Objects and Action Likelihoods

In this paper we propose a method which can improve MRF-Particle filters used to solve the hijacking problem; independent particle filters for tracking each object can be kidnapped by a neighboring target which has higher likelihood than that of real target. In the method the motion model built by Markov random field (MRF) has been usefully applied for avoiding hijacking by lowering the weight of particles which are very close to any neighboring target. The MRF unary and pairwise potential functions of neighboring targets are defined as the penalty function to lower the particle’s weights. And potential function can be reused for defining action likelihood which can measure the motion of object group. C. Oh, C. Lee, Tracking Multiple Objects and Action Likelihoods, 14th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD 2013), SCI 492, 229-240, 2013