Multi-target tracking via multiple cost-reference particle filtering

Abstract

In this paper we address the problem of multitarget tracking in a network of sensors collecting received signal strength measurements. In order to deal with the nonlinear nature of the system, we apply the particle filtering methodology. The focus is on high-dimensional systems, i.e., scenarios with large number of targets. This justifies the use of an interconnected bank of particle filters. At each algorithmic step, each individual particle filter tracks one target, thereby minimizing the load of each filter. The filters need to send/receive the necessary information to/from other filters for correct functioning and accurate performance. The individual filters do not use any probabilistic assumption about the noises in the system in order to obtain a more robust scheme. Alternatively, they employ a user-defined cost function, which makes the resulting method more flexible. Computer simulations show the validity of the approach and reveal a good performance of the proposed method when compared to existing techniques.

Bayesian filtering based on solving the Fokker-Planck equation

“…Nonlinear filtering has many applications in science and engineering fields, such as target tracking (Bugallo, 2015), navigation (Valli et al, 2014), attitude determination (Li et al, 2014), rendezvous and docking (Karlgaard, 2006). In these fields, the nonlinear filtering can be used to optimally estimate the state of dynamic system when the measurements become available.…”