A Hybrid Labeled Multi-Bernoulli Filter With Amplitude For Tracking Fluctuating Targets

The amplitude information of target returns has been incorporated into many tracking algorithms for performance improvements. One of the limitations of employing amplitude feature is that the signal-to-noise ratio (SNR) of the target, i.e., the parameter of amplitude likelihood, is usually assumed to be known and constant. In practice, the target SNR is always unknown, and is dependent on aspect angle hence it will fluctuate. In this paper we propose a hybrid labeled multi-Bernoulli (LMB) filter that introduces the signal amplitude into the LMB filter for tracking targets with unknown and fluctuating SNR. The fluctuation of target SNR is modeled by an autoregressive gamma process and amplitude likelihoods for Swerling 1 and 3 targets are considered. Under Rao-Blackwell decomposition, an approximate Gamma estimator based on Laplace transform and Markov Chain Monte Carlo method is proposed to estimate the target SNR, and the kinematic state is estimated by a Gaussian mixture filter conditioned on the target SNR. The performance of the proposed hybrid filter is analyzed via a tracking scenario including three crossing targets. Simulation results verify the efficacy of the proposed SNR estimator and quantify the benefits of incorporating amplitude information for multi-target tracking.