On particle filters in radar target tracking

The dissertation focused on the research, implementation, and evaluation of particle filters for radar target track filtering of a maneuvering target, through quantitative simulations and analysis thereof. Target track filtering, also called target track smoothing, aims to minimize the error between a radar target's predicted and actual position. From the literature it had been suggested that particle filters were more suitable for filtering in non-linear/non-Gaussian systems. Furthermore, it had been determined that particle filters were a relatively newer field of research relating to radar target track filtering for non-linear, non-Gaussian maneuvering target tracking problems, compared to the more traditional and widely known and implemented approaches and techniques. The objectives of the research project had been achieved through the development of a software radar target tracking filter simulator, which implemented a sequential importance re-sampling particle filter algorithm and suitable target and noise models. This particular particle filter had been identified from a review of the theory of particle filters. The theory of the more conventional tracking filters used in radar applications had also been reviewed and discussed. The performance of the sequential importance re-sampling particle filter for radar target track filtering had been evaluated through quantitative simulations and analysis thereof, using predefined metrics identified from the literature. These metrics had been the root mean squared error metric for accuracy, and the normalized processing time metric for computational complexity. It had been shown that the sequential importance re-sampling particle filter achieved improved accuracy performance in the track filtering of a maneuvering radar target in a non-Gaussian (Laplacian) noise environment, compared to a Gaussian noise environment. It had also been shown that the accuracy performance of the sequential importance re-sampling particle filter is a function of the number of particles used in the sequential importance re-sampling particle filter algorithm. The sequential importance re-sampling particle filter had also been compared to two conventional tracking filters, namely the alpha-beta filter and the Singer-Kalman filter, and had better accuracy performance in both cases. The normalized processing time of the sequential importance re-sampling particle filter had been shown to be a function of the number of particles used in the sequential importance re-sampling particle filter algorithm. The normalized processing time of the sequential importance re-sampling particle filter had been shown to be higher than that of both the alpha-beta filter and the Singer-Kalman filter. Analysis of the posterior Cramér-Rao lower bound of the sequential importance re-sampling particle filter had also been conducted and presented in the dissertation.

Grantor dc:publisher.institution

Department of Electrical Engineering

Year dc:date.issued

2016

Author dc:creator

• Bauermeister, Etienne F

Advisor dc:contributor.advisor

• O'Hagan, Daniel

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/11427/22787

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/22787