Joint detection and tracking of moving targets in clutter

This thesis develops an integrated framework for detection and tracking of targets that move randomly in clutter. We introduce models for the target motions, target signatures, and clutter in a variety of scenarios: one dimensional (1D) and two dimensional (2D) surveillance spaces; pointwise and extended targets with floating and drifting motion; single and multiple targets; deterministic and random signatures; white and spatially correlated Gaussian clutter; and non-Gaussian clutter with heavy-tail statistics (K and Weibull envelopes).; We derive the optimal nonlinear multitarget detector/tracker and develop efficient implementations that reduce the computational complexity of the algorithm by several orders of magnitude. We also propose suboptimal implementations that are based on approximations of the actual structure of the optimal detector/tracker rather than ad-hoc assumptions. These suboptimal implementations further simplify the computations while retaining good performance.; We present receiver operating characteristic curves (ROC) for detection of targets in white and correlated Gaussian clutter, and in white non-Gaussian clutter. We compare the tracking behavior with other commonly used trackers, like the peak detector and the extended Kalman-Bucy filter (KBf). Our performance studies show that there is a significant margin for improvement over existing approaches found in the literature.