Robust Estimation for Target Tracking Using a Multiple Synthetic Filter Approach

A Multiple Synthetic Filter Approach (MSFA) is proposed to perform target tracking for a given benchmark problem of highly maneuvering targets. The proposed MSFA, which is synthetically designed using the neural filtering techniques to produce an optimal synthetic filtering scheme as opposed to the traditional analytic Kalman filtering approach, possesses the following attractive performance improvements: (1) it provides robust estimation in the presence of system uncertainties (e.g., mismatch between system and filter model, measurement noise variations, etc.); (2) it outperforms the extended Kalman filter and the iterated Kalman filter subject to nonlinearity; (3) it maintains target track when the target drastically changes its acceleration (target maneuverability variations); and (4) it provides a most suitable form for real-time filtering implementation in a radar tracker due to the massively parallel nature of the neural network filtering architecture. Design issues such as sampling rate, frequency of interest, and their related design criteria will also be investigated to insure that the proposed MSFA will produce improved estimation performance.