Skip to Main Content
When compared to the tracking problem in which prior knowledge is available, generating the initial distribution for the state vector of a phenomenon of interest, with no prior knowledge of the desired state, is a challenging problem. In this paper, the authors develop a fully distributed initialization algorithm that fuses data in heterogeneous sensor networks using communication trees. Monte Carlo methods are used to fuse the collected data and to represent the desired state vector distribution. The presented algorithm utilizes an importance function that is additive in the local node posterior distributions, providing a robust alternative to belief propagation methods in which particles are generated according to the product of local node posteriors.