Sequential Monte Carlo (SMC) methods for Bayesian inferencernhave been applied to the problem of informationdrivenrndynamic sensor collaboration in clutter environmentsrnfor sensor networks. The dynamics of the system under considerationrnare described by nonlinear sensing models withinrnrandomly deployed sensor nodes. The exact solution to thisrnproblem is prohibitively complex due to the nonlinear naturernof the system. The sequential Monte Carlo (SMC)rnmethods are therefore employed to track the probabilisticrndynamics of the system and to make the corresponding Bayesianrnestimates and predictions. To meet the specific requirementsrninherent in sensor network, such as low-power consumptionrnand collaborative information processing, we proposerna novel SMC solution that makes use of the auxiliaryrnparticle filter technique for data fusion at densely deployedrnsensor nodes, and the collapsed kernel representation of therna posteriori distribution for information exchange betweenrnsensor nodes. Furthermore, an efficient numerical methodrnis proposed for approximating the entropy-based informationrnutility in sensor selection. It is seen that under the SMCrnframework, the optimal sensor selection and collaborationrncan be implemented naturally, and significant improvementrnis achieved over existing methods in terms of localizing andrntracking accuracies.
展开▼
