Signal processing techniques for multiple target localisation and tracking

Abstract

Target geolocation and tracking systems have a wide spectrum of important practical applications ranging from military to biomedical systems. This thesis explores two main areas of target geolocation and tracking systems: (i) the active target localisation of multiple moving targets using wideband chirp signals received at a sensor array, for both the far and near field cases; and (ii) the passive tracking of a single target moving with nearly constant velocity using the sequence of bearing measurements received at a single observation platform, also known as the bearings-only tracking problem. It is challenging for traditional parameter estimation methods to effectively localise multiple moving targets using wideband signals received at a sensor array due to the frequency dependent nature of the array steering vector. We develop a technique for processing the wideband signals to eliminate the frequency dependency in the array steering vector so that existing high-resolution parameter estimation methods designed for narrowband signals can be applied for the localisation of both far and near field targets. As chirp signals are used as wideband signals in our analysis, the properties of the fractional Fourier transform were exploited to improve the accuracy and resolution of the target localisation system. The performance of most existing nonlinear Bayesian filtering algorithms in highly nonlinear scenarios of the bearings-only tracking problem is unsatisfactory. The particle filter can achieve excellent performance but it is computationally expensive. We present an alternative, efficient nonlinear filtering algorithm, called the Gaussian-sum cubature Kalman filter for the bearings-only tracking problem and it demonstrates comparable performance to the particle filter with significantly reduced computational cost. The performance of the proposed algorithm can be improved by smoothing the estimated states using future measurements and we present Gaussian-sum cubature Kalman smoothers for the bearings-only tracking problem. In addition to that, we develop a track-before-detect algorithm for simultaneously detecting and tracking a single target to tackle scenarios with low signal-to-noise ratio conditions.