Abstract:
The work presented in this thesis focuses on the problem of reconstructing threedimensional models of fire from real images. The intended application of the reconstructions is for use in research into the phenomenon of fire-induced high voltage flashover, which, while a common problem, is not fully understood. As such the reconstruction must estimate not only the geometry of the flame but also the internal density structure, using only a set of a few synchronised images. Current flame reconstruction techniques are investigated, revealing that relatively little work has been done on the subject, and that most techniques follow either an exclusively geometric or tomographic direction. A novel method, termed the 3D Fuzzy Hull method, is proposed, incorporating aspects of tomography, statistical image segmentation and traditional object reconstruction techniques. By using physically based principles the flame images are related to the relative flame density, allowing the problem to be tackled from a tomographic perspective. A variation of algebraic tomography is then used to estimate the internal density field of the flame. This is done within a geometric framework by integrating the fuzzy c-means image segmentation technique and the visual hull concept into the process. Results are presented using synthetic and real flame image sets.
