Title:  Use of Gaussian Mixture Distribution Models for Address Non-gaussian Errors in Radar Target Tracking

Committee: 

Dr. Verriest, Advisor

Dr. Blair, Co-Advisor   

Dr. Egerstedt, Chair

Dr. Lanterman

Abstract:

The objective of the proposed research is to develop a computationally efficient solution to the problem of processing radar measurements with highly accurate range information compared to other dimensions. When converting to Cartesian form, the measurement error distribution becomes highly non-Gaussian.  Existing methods using a single Gaussian to represent the measurements suffer from inconsistent error covariance reporting and/or loss of performance in the range dimension, whereas particle filter approaches have a high computational complexity. Therefore, the Cartesian radar measurement errors will be modeled by a distribution consisting of a weighted sum of Gaussian densities, known as a Gaussian mixture. The parameters for the Gaussian mixture components will be chosen based upon a library generated using efficient numerical methods such as the Expectation Maximization algorithm rather than ad-hoc methods present in existing work. With this measurement model, a Gaussian mixture Kalman Filter will be implemented with the goal of achieving efficient computational complexity, consistent covariance reporting, and accurate range estimation performance, while limiting the number of Gaussian components used to represent the measurements and track.  Both monostatic and bistatic radar measurements will be considered.