| Standard models used in target tracking algorithms postulate a sensor system which delivers information on the state dynamics of the center of mass of the target (positions, velocities, accelerations, etc.). A radar is such a sensor. However, new classes of sensors provide information which is not directly related to motion of the target center of mass, but which is related to important aspects of the problem. A prime example is an imaging sensor.; An imaging sensor, by capturing information about the physical extent of the target on the image plane, conveys information about its orientation in space, and about its identification. Identification adds flexibility in that it permits the selection of the "best" target. Orientation can give important information on target intentions, in terms of identifying the most likely directions of maneuvering accelerations. This information, when coupled with the traditional measurement of the dynamics of the center of mass, provide for more flexible and accurate tracking systems.; This research concentrates on two aspects of such a tracking system. The first is the development of an appropriate model for the measurement information flowing into the system from the imaging sensor. This model must not only capture the unique features of the imager, but also be mathematically tractable in the context of the larger problem. This research proposes to characterize this information as a marked point process. The second aspect considered here is the characterization of the features and performance of such a filter. This requires identifying the important parameters associated with this filter, and studying the effect of their variations on filter performance.; This research inquiry concludes with the study of a specific tracking scenario, in which a traditional sensor and an imaging sensor are combined to produce a system which is considerably more accurate than the traditional sensor acting alone. |
