Multiple target tracking and classification using the unscented probability hypothesis density filter

Tracking an unknown and time varying number of targets is a difficult issue. The Unscented Probability Hypothesis Density Filter (UKPHD) tackles this problem, moreover, it allows the estimation of both the number of targets and their states without any data association steps by considering the target states as a single global target state, its a closed-form solution for the probability hypothesis density (PHD) filter that deals with non linear systems, it propagates the first-order moment of the multitarget posterior instead of the posterior distribution itself because evaluating the multiple-target posterior distribution is currently computationally intractable for real-time applications in multiple Target tracking problems. However, targets are poorly described by a single dynamic model, in fact, they may change their kinematic model at any time which makes the tracking algorithm incapable of estimating efficiently the true trajectories. The Interacting Multiple Model (IMM) algorithm is used to address this. The IMM uses multiple models to describe targets behavior and adaptively determines which model(s) are the most appropriate at each time step. In this paper, we present a new interacting multiple model Unscented probability hypothesis density filter (IMM-UKPHD) to deal with the problem of tracking a time varying number of maneuvering targets. In our approach, a bank of Unscented probability hypothesis density filters is used in the interacting multiple model (IMM) framework for updating the state of moving targets. Simulation results show the efficiency of the proposed algorithm.