The Unscented Kalman Particle PHD filter for joint multiple target tracking and classification

The probability hypothesis density (PHD) is the first order statistical moment of the multiple target posterior density; the PHD recursion involves multiple integrals that generally have no closed form solutions. A (Sequential Monte Carlo)SMC implementation of the PHD filter has been proposed to tackle the issue of joint estimating the number of targets and their states. However, because the state transition does not take into account the most recent observation, the particles drawn from prior transition may have very low likelihood and their contributions to the posterior estimation become negligible. In this paper, we propose a novel algorithm named Unscented Kalman Particle PHD filter (UKP- PHD), and associate it with Multiple dynamical Models (MM)method. The algorithm consists of a P-PHD filter that uses an Unscented Kalman filter to generate the importance proposal distribution; the UKF allows the P-PHD filter to incorporate the latest observations into a prior updating routine and thus, generates proposal distributions that match the true posterior more closely. Moreover, The MM solves the problem of tracking manoeuvring targets. Simulation shows that the proposed filter outperforms the P-PHD filter.