Arresting Particle Hijack in Multitarget Tracking

Robust, lightweight, and distributed multitarget tracking in wireless sensor networks (WSN) is discussed. Sequential Monte Carlo or particle filter is an excellent stochastic technique for approximate state estimation using Bayesian algorithms. When targets move close to each other, the signal energies generated by them get overlapped. Solving multitarget tracking problem using independent particle filters suffers from many problems, including that the target with the best likelihood hijacks the particles of nearby targets. Even in joint tracking there is potential for particles to carry labels to wrong target position. This paper proposes a dynamic-Bayesian-network-based framework called ‘Particle Filter-Merger Modeled-Maximum a Posteriori’ (PF-MM-MAP). The framework explicitly models the signal overlap by introducing additional hidden states, thereby adding robustness to the tracking process. The implementation is distributed and lightweight, keeping in view the requirements and resource constraints of WSN.