Soft Computing Technique (PLE) for Target Tracking

The pseudo linear Kalman filter for target tracking concerns the estimation of target motion parameters i.e., range, bearing, course and speed of a moving target, from noisy corrupted data. In the ocean environment. two dimensional bearings-only target motion analysis is generally used. An observer monitors sonar bearings from radiating target in passive listening mode. An observer processes these measurements and finds out target motion parameters like, range, course, bearings and speed of the target. As range is not available and the bearing measurement is not linearly related to the target states, the whole process becomes nonlinear. Added to this, since bearing measurements are extracted from passive sonar, the process remains unobservable until observer executes a proper maneuver. The measurements are corrupted with noise, hence the process becomes the random process. The pseudo linear filter is projected in such a way that it does not require any initial estimate at all and at the same time offers all the features of the extended kalman filter based

pseudo-linear filter; namely sequential processing, flexibility to adopt the variance of each measurement. The algorithm is tested in Monte Carlo simulations and results are presented for one typical scenario. Effect of random noise in the range ,course and speed distribution is presented.

INTRODUCTION

The basic problem in target motion analysis (TMA) is to estimate the trajectory of an object (i.e., position and velocity) from noise corrupted sensor data. In the context of ocean environment, the moving observation platform (I.e., receiver, observer, own ship) passively monitors sonar bearings from a single acoustic source (i.e., target),which is assumed to be traveling with uniform velocity. The observer processes these measurements and finds the target motion parameters viz., range, course, bearing and speed of the target.

The measurement process is nonlinear, because the range measurement is not available and the Learning measurement is not linearly related to the target state, making the whole process as non-linear. Since the bearing measurements are extracted from single sonar, toe process remains unobservable until the observer executes a proper maneuver. In order to make the process observable, the observer should execute with better bearing rate. To achieve the better bearing rates, the observer's trajectory is composed of constant velocity segments termed as legs. The TMA process is not completely observable for any single leg. Although two distinct legs pennit a unique solution, the degree of convergence attained on a given leg is restricted and several legs are required to achieve an acceptable error. ,In addition to the measurement noise, the performance of any bearings only TMA estimation technique IS affected by the geometric characteristics of the observer's maneuver strategy. Several estimation techniques have been applied to the bearings only Target Motion Analysis problem with varying results. When implemented in Cartesian state space, the Extended Kalman Filter exhibits divergence problems. 

Although the EKF may often yield good estimates, itcan in many instances diverge, yielding poor estimates, An alternative method, based on pseudo-measurements, which are derived from the known observer state and available measured bearing, and it is linearly related to the target state referred to as pseudo linear estimate (PLE). However ityie1ds some reliable estimates, the PLE is known to produce a biased state estimate, due to the liberalization of the process. The process of extracting useful information from a signal and discarding the extraneous is call signal processing. This project is concerned with the implementation of signal processing techniques to extract pertinent signal information from random signals utilizing any priory information available. We call these techniques as signal estimation, and we call the particular algorithm a signal estimator or just estimator. Some times estimators are called as filters (e.g.Kalman filter) because they perform the same function as a deterministic, filter except for random signals i.e., they remove unwanted disturbances. The estimator to produce 'filtered date' processes noisy

measurements.