Feature-Aided UKF-PHD Filter for Nonlinear Multi-target Tracking in Holographic Staring Radar

Traditional radar multi-target tracking relies on data association methods, which result in exponential computational complexity when clutter and target densities are high. The Gaussian mixture probability hypothesis density (GM-PHD) filter, built upon the random finite set framework, presents a more efficient approach for multi-target tracking without having to conduct explicit data association. However, this method is only applicable to Gaussian linear models, limiting its utility in practical scenarios. Moreover, radar environments often feature high clutter densities and low detection probabilities, making real-time and accurate target tracking challenging. To address these issues, this paper proposes extending the GM-PHD filter to nonlinear Gaussian models using the unscented transform and introduces a feature-aided unscented Kalman filter PHD (FAUKF-PHD) multi-target tracking algorithm. By leveraging the high-precision Doppler information contained in holographic staring radar echoes, the proposed algorithm adjusts the weights of Gaussian components, significantly reducing clutter weights to distinguish targets from clutter and pruning redundant components. The simulation outcomes evidently show that the FAUKF-PHD filter put forward in this study boosts the estimation accuracy and real-time performance under scenarios characterized by high clutter density and low detection probability.