Hybrid Particle Filter Based Dynamic Compressed Sensing for Signal-Level Multitarget Tracking

We propose a novel algorithm, state propagation based dynamic compressed sensing (SP-DCS), that uses a target dynamic model in dynamic compressed sensing (DCS) to track a fixed number of targets. To track a time-varying number of targets using raw measurements from a Doppler radar, we also propose a novel hybrid particle filter based dynamic compressed sensing (HPF-DCS) algorithm. We calculate the support set in a Bayesian framework and a particle filter approximates the posterior probability mass function (pmf) of the support set. HPF-DCS is a combination of random and deterministic sampling. In random sampling, a number of predicted existing sub-particles are sampled from the prior pmf of the existing support set to handle the scenario when targets disappear randomly at a scan time. In deterministic sampling, the new support set corresponding to newly appearing targets is calculated by solving a sparsity promoting optimization problem. Our simulation results show that the proposed algorithm can track a time-varying number of targets successfully. It also outperforms the sequential Monte Carlo based probability hypothesis density (SMC-PHD) filter, as well as the multi-mode, multi-target track before detect (MM-MT-TBD) filter.