Particle filtering methods for incorporating intelligence updates

Abstract

Due to uncertainty in target locations, stochastic models are implemented to provide a representation of location distribution. The reliability of these models has a profound effect on the ability to successfully interdict these targets. A key factor in the reliability of a model is the incorporation of information updates. A common method for incorporating information updates is Kalman filtering. However, given the probable nonlinear and non-Gaussian nature of target movement models, the fidelity of solutions provided by Kalman filtering could be significantly degraded. A more robust methodology needs to be employed. This thesis uses an updating algorithm known as particle filtering to incorporate information updates concerning the target's position. Particle filtering is a nonparametric filtering technique that is adaptable and flexible. The particle filter is incorporated into a model that uses a stochastic process known as a Brownian bridge to model target movement. A Brownian bridge models target movement with minimal information and allows for uncertainty during periods when target location is unknown. As new intelligence arrives, the particle filter is used to update a probabilistic heat map of target position. The main goal of this thesis is to design a stochastic model integrating both the Brownian bridge model and particle filtering.