| Two primary topics are addressed in this dissertation. The general problem that is examined is the detection and tracking of small, dim, moving targets in a series of digital images, with an emphasis on airborne targets in satellite imagery. Both two-dimensional and three-dimensional approaches to this problem are considered, and their detection and location estimation performance is compared.; The second major thrust of this dissertation is an analysis of the Hough transform, which is sometimes used in the target detection and tracking problem, and a comparison of it to an optimal processor derived using statistical signal detection theory. The relationship of these two processors is identified, and their detection and location estimation performance is evaluated, both analytically and using computer simulation, and compared as a function of signal-to-noise ratio in several different noise environments. Furthermore, the proper resolution of the Hough transform parameters, {dollar}rho{dollar} and {dollar}theta{dollar}, under various conditions is analyzed in detail.; In general, the signal detection theory processor performs significantly better than the Hough transform. This is due to several refinements it makes to the Hough transform algorithm. Both processors are found to be robust with respect to noise environment. Both processors can be negatively affected by coarse quantizations of {dollar}rho{dollar} and {dollar}theta{dollar}, especially at high signal-to-noise ratios, but several approaches to mitigating the negative effects are derived.; A three-dimensional approach to the target detection and tracking problem, in which the series of frames is not projected onto a plane, is found to yield dramatically higher performance than the two-dimensional projection approach, due to the much higher signal-to-noise ratio of the three-dimensional data. It is also shown that in an ongoing process in which a recursive updating scheme can be used, the computational complexity of the three-dimensional processor is only about twice that of the two-dimensional processor. |
