Multi-sensor multi-target tracking and track association is a research topic with applications in many areas, including radar and sonar systems, and has received considerable and continuous attention in the literature since the early 70's. The two key problems that researchers are concerned with are those of measurement-to-target association and state-estimation. Of the two, the thorny problem is that of measurement-to-target association. In essence, the association task is to find optimum techniques that will assign a received set of measurements to a set of target models or to clutter. Once the association is known standard estimation techniques can be used to obtain the target state estimates. Current conventional multi-target tracking and track association methods rely on sub-optimum data association solutions to assign measurements to target models. Current optimum multi-hypothesis algorithms rely on enumeration and are computationally prohibitive, resulting in pruning of all possible solutions to the most likely subset, thus again becoming suboptimum.
This dissertation is concerned with the development of optimum multi-sensor multi-target tracking and track association techniques. First the Probabilistic Multi Hypothesis Tracking (PMHT) algorithm, a recently developed multi-target tracking algorithm [1, 2], is extended to work with multiple sensors. As in the original PMHT, the extended algorithm treats the estimation and association problems jointly and seeks an optimum solution. The resulting algorithm does not require enumeration and pruning, and works on the full set of available measurements. As part of this work, a multi-sensor multi-target Cramer Rao Lower Bound (CRLB) based on the marginal PMHT density is derived and used to evaluate performance bounds of the resulting algorithm, relative to an optimum single target lower bound. Next, a joint track-to-track association and estimation algorithm in clutter, the Track Segment PMHT (TSPMHT), is developed. This is the key contribution of the dissertation, since this is one of the first reported methods in the literature that deals with an optimum probabilistic solution for multiple track segment association in clutter, especially for track segments with varying lengths. The proposed algorithms are applied to problems in the field of sonar and their performance is evaluated through simulation. Finally the last section of this thesis deals with the performance analysis of track association tests under various Signal to Noise Ratio (SNR) conditions.
