Improving radar target tracking with the range rate measurement

The radar tracking problem comprises measurement, association, and filtering. Radar measures a target's range, bearing, and range rate, which is target speed along a line extending from the radar to the target. Radar cannot directly measure acceleration, which is why the tracking problem is hard. The tracking system associates measurements to nearby, existing tracks. Unassociated measurements can start new tracks; associated measurements update the current track state estimates through Kalman filters. Our research addresses the filtering part of the tracking problem.; Although association algorithms frequently consider range rate, tracking filters generally ignore it. The reason is simple: range rate is highly nonlinear in a Cartesian coordinate system, and so it is unsuited for Kalman filters. Linearized approximations of range rate are used but the results are sometimes unsatisfactory. This is unfortunate, because range rate is the only measurement of target velocity. Filters that ignore range rate, consequently, are not using all of the available information.; Interacting multiple models, a robust derivative of the Kalman filter, are the most common, maneuvering target tracking filter. Recently, there has been interest in other types of robust filters but many of these have not yet been applied to the tracking problem. The newer filters are more complicated; however, the performance gains may be worth the additional computations. Until now, there have been no systematic efforts to evaluate the applicability of modern, robust filters to the tracking problem.; Our research has three parts. The first two parts, which are intended to better use the range rate measurement, are the discoveries of a new range rate linearization, and a mapping from range rate into a statistic of accelerations. Analytical and numerical analyses of these discoveries confirm that they can improve target trackers. In the third part of our research, we systematically compare interacting multiple models to other robust filters, with the goal of determining whether other robust filters could benefit target tracking systems.