Adaptive noise covariance PHD filter under nonlinear measurement

Probability hypothesis density (PHD) filter has been demonstrated to be an effective approach for multi-target tracking in real time. However, these methods based on the PHD filter assume that the measurement noise covariance is known as a priori. This is unrealistic for real applications because it may be previously unknown or its value may be time-varying as the environment changes. To solve this problem, an adaptive noise covariance algorithm for multi-target tracking under the nonlinear measurement is proposed. Based on the PHD filter, the proposed algorithm employs the cubature Kalman (CK) technology to approximate the nonlinear model, models the noise covariance distribution as inverse Wishart (IW) distribution, and recursively estimates the joint posterior density of the measurement noise covariance and multi-target states by the variational Bayesian (VB) approach. The simulation results indicate that the proposed algorithm could effectively estimate measurement noise covariance, and achieve the accurate estimation of the target number and corresponding multi-target states.