Gaussian Process Methods for Target Tracking and Sensor Scheduling

This thesis explores the application of Gaussian Processes to practical target tracking
problems. It begins with the development of a general-purpose codebase for implementing
Gaussian Processes, creating a flexible foundation for various tracking scenarios. To tackle
the computational challenges inherent to Gaussian Processes, the research investigates
several sparse Gaussian Process methods. These methods reduce computation time while
preserving accuracy, demonstrating significant improvements in both inference speed and
tracking performance.
The study advances these foundational methods by integrating variance-based Bayesian
optimisation techniques to enhance tracking and sensor management for Unmanned Aerial
Vehicles. This approach utilises the probabilistic nature of Gaussian Processes to optimise-
mize resource deployment dynamically, boosting target localisation and tracking efficiency.
Beyond continuous tracking, the research extends to Gaussian Process Classification to
address discrete decision-making scenarios. These additions highlight the versatility of
Gaussian Processes in Target Tracking Applications.
The results confirm that Gaussian Processes, when combined with sparse approximations-
tions and Bayesian optimisation, offer an effective and scalable framework for improving
target tracking performance. By focusing on dynamic and adaptive methods, this research
contributes to the development of tracking algorithms that meet the demands of diverse
and evolving scenarios.