Initialization of Multi-bernoulli Random-finite-sets over a Sensor Tree

We study a framework of multi-target state initialization employed over a tree-like sensor network. Such network provides graceful aggregation of target state statistics by exploiting limited bandwidth, and makes it possible to approximate the target states based on particle filtering defined in the context of Random-FiniteSet (RFS) theory. This contrasts with conventional multi-target filtering approach based on RFS over a star topology. In the tree structure, the root sensor node launches the initialization and passes messages downward to its leaf sensors. Each leaf sensor node collects upward messages from its subsidiary nodes and updates target state statistics. We implement the initialization process through the Sequential Monte-Carlo (SMC) method, which requires the sampling and re-sampling steps different from that in a non-RFS type initialization and aggregation.