University of Houston

This dissertation describes a computational method that accurately tracks the image of tissue undergoing non-rigid deformations as it moves. It involves multiple particle filter trackers that work together in a collaborative way. Each particle filter tracker is driven by a probabilistic template function with both spatial and temporal smoothing components, which is capable of adapting to abrupt positional and physiological changes. The trackers' spatio-temporal interactions are modeled via a dynamic Bayesian network. This tracker network achieves robust real-time performance over long periods of time. It is a general methodology, applicable across imaging modalities and in a variety of applications, including thermophysiological measurements on the face and MRI-guided robotic interventions on the heart. An assortment of experiments brings up the method's potential.