Hybrid Particle/Extended Finite Memory Filter to Improve Target Tracking Accuracy of Radar Measurement in Harsh Environments

Abstract

Although the particle filter (PF) generally provides accurate estimation, it fails in harsh environments, such as severe signal noise and/or abrupt state change. The PF also requires a number of particles for accurate estimation, causing heavy computational burden. Therefore, it is difficult to use the PF for real applications. To overcome PF drawbacks, we propose the extended finite memory (EFM) filter and hybrid particle filtering algorithm combining the regularized particle filter (RPF) as the main filter with an auxiliary EFM filter. The hybrid particle/EFM filter can detect RPF failure and reset the particles using an EFM estimation. The proposed filter shows robust performance against severe signal noise and abrupt change of target motion. Experiments using vehicle radar signals were performed in harsh environments to compare the proposed tracker with current best practice regularized particle and extended Kalman trackers.