Multitarget Distributed Tracking With Probability Hypothesis Density in Large-Scale Sensor Networks

In this article, multitarget distributed tracking problems in large-scale sensor networks consisting of local fusion centers (LFCs) and sensor nodes (SNs) are considered. Each SN transmits measurements to its superior LFC, which runs a local probability hypothesis density filter having Gaussian mixture representation (GM-PHD) using measurements from subordinate SNs. A framework based on random finite set (RFS) is developed, incorporating an improved GM-PHD filter and an improved geometric average (GA) fusion rule. Regarding the local GM-PHD filter, a modified adaptive birth model using presegmented measurements and an SN-related clutter-based update step are proposed to handle the measurement origin uncertainty and the cardinality overestimation issues. Regarding the proposed distributed fusion algorithm that fuses posteriors among LFCs with the proposed GM-PHD filters, a hybrid fusion rule is proposed to compensate the GA miss detection by arithmetic average (AA) fusion. The boundedness of the proposed algorithm is derived. Finally, simulations illustrate the effectiveness of the proposed algorithm.