A Particle filter for mutistatic radar tracking

Abstract

This paper introduce a new Automatic radar tracking system with multiple observations from Multistatic radar using a Particle filter as a nonlinear predictor for data fusion and prediction. The algorithm is based on using Particle filter instead of using a linear or non-linear Kalman Filter (Extended Kalman Filter). Particle filtering, also known as Sequential Monte Carlo, is an attractive estimation procedure for non-linear dynamical systems. Recently several popular methods such as Forward Backward Smoother (FBS), and Maximum A-Posteriori Smoother (MAP) have been introduced into the literature. These techniques involve a re-computation of the discrete distribution obtained from the Particle filter. While the smoother offers an improvement in the estimation, there is a significant computation cost that often makes this step unattractive in practice. The system is simulated using Matlab to compare the performance of the estimation routines of both the Kalman and Particle filters, and Particle filter without and with smoothers. The processing time is also studied. Simulation results shows that the Kalman filter improves the automatic tracking performance with multiple observations. Particle filter improves the fusion and prediction estimate of the non-linear moving object in presence of measurement errors.