Dynamic Editable Models of Fire From Video

Abstract

This thesis presents the following statement: it is possible to generate three-dimensional, intuitively editable, dynamic models of flames from video. Current techniques allow flames to be modelled via physical simulations; but these methods require significant effort to produce content, and due to unintuitive parameters it is difficult to produce a specific visual result. Alternatively, models can be acquired from video; but only as a density field representation, and this is not editable. In this document, a novel method for modelling flames is introduced, the ‘flame core model’. This is approach consists of two parts: a structural element – the core – and the density that surrounds it. The parameters of this model are intuitively editable and can be found from three-dimensional video data. Using these, new flames can be generated from existing sequences. This significantly simplifies the problem of content generation – from the laborious task of a skilled user, to the simple act of filming video of a real flame. Some flame appearance qualities can also be learned from simple examples. With this technique, it is possible to use a single frame of a single uncalibrated camera to provide parameters for the appearance of a three-dimensional moving flame, something that is completely novel for flame reconstruction literature. The video data filmed for this project and the three-dimensional reconstructions are also state of the art for the field.

Date of Award	29 May 2015
Original language	English
Awarding Institution	University of Bath
Supervisor	Peter Hall (Supervisor)