Artists in the films and games industry often want to make models of natural phenomena such as tree, water, smoke, fire. This is no easy task: the models are often designed by engineers and are physically based, which makes them hard for artists to use; artists sometimes want the models to behave non-physically, which gives the engineers problems in designing an underlying model; the process is difficult and time consuming and so is expensive. This talk describes how to acquire three-dimensional, dynamic models of complex natural phenomena from a single consumer level video camera. The default output looks natural but is easy to edit to behave in the way artists want. We will discuss trees, water, and steps towards a solution for more general phenomena.
Peter is Professor in the Visual Computing Group and is Director of the Centre for Visual Entertainment. His research spans more than twenty years is primarily directed at style transfer, the acquisition of complex moving 3D models from video, and recognition; additional interests include machine learning, robotics, and accessible computing. He works with companies both large and small including Disney, Ninja Theory, dcActive and Lego. He has international academic collaborations in Netherlands, Germany, Canada, and China. Peter is Associate Editor with three journals; he has published in SIGGRAPH, SIGGRAPH Asia, CVPR, ICCV, ECCV, ICRA, TVCG, TIP, PAMI and many others.