Algorithms for multisensor maneuvering target tracking

Target tracking is the processing of sensor measurements (e.g., radar returns) obtained from a target (moving objects such as airplanes, tanks or submarines) in order to maintain an estimate of its state (position, velocity and acceleration). In this dissertation, a set of novel algorithms are developed for multisensor maneuvering target tracking. The main emphasis in this dissertation is on fixed-lag smoothing where we estimate the target state at time k − d given measurements up to time k ( d > 0). (1) We propose a novel interacting multiple model (IMM) fixed-lag smoothing algorithm for Markovian switching systems. Using a state augmentation approach, we transform the smoothing problem for the original system into a filtering problem for a state-augmented system where IMM filtering algorithm can be applied. We then apply the algorithm to tracking a maneuvering target since the behavior of a maneuvering target can be described by a set of hypothesized models and a maneuver can be modeled as a switching from one model to another model. (2) We want to improve the current state estimate of the target by introducing an IMM fixed-lag smoothing based filtering algorithm. Fixed-lag smoothing with delay d provides target state estimate at time k − d given measurements up to time k. In filtering we are interested in state estimate at time k. The main idea of our novel algorithm is to switch between an IMM filter and a single maneuver model filter by detecting onset and termination of maneuvers. This is achieved by examining the smoothed model probabilities. (3) We extend our basic IMM smoothing algorithm to a more complicated tracking scenario, tracking a target in clutter. An existing probabilistic data association (PDA) approach has proved to be an effective way to solve the problem of measurement origin uncertainty. We combine the basic IMM fixed-lag smoothing algorithm with the PDA approach to develop an IMMPDA fixed-lag smoothing algorithm for tracking a maneuvering target in clutter with multiple sensors. (4) We consider the problem of tracking multiple targets. The measurement-to-target association problem is even more complicated in this case. A measurement may either originate from a target of interest, or clutter, or a neighboring target. An existing joint probabilistic data association (JPDA) approach provides an effective solution to this problem provided that the number of targets are known. We develop an IMM/JPDA fixed-lag smoothing algorithm for multiple maneuvering target tracking with multiple sensors by combining the basic IMM fixed-lag smoothing algorithm with the JPDA approach. The simulation results show that all our proposed IMM smoothing algorithms offer much better performance compared to that achieved by the corresponding IMM filtering algorithms. (5) Finally, we have made some preliminary efforts in tracking targets based on passive sensors. Passive sensors are mainly used to track the direction of arrival (DOA) of targets. We investigate two existing higher-order statistics based blind source separation algorithms. In addition we also investigate a second-order statistics based DOA tracking algorithm to offer a performance comparison. (Abstract shortened by UMI.)...