Data driven REActive Modelling

点击翻译按钮获取中文摘要

Data Driven Reactive Modelling: Summary We live in a world where computer graphics plays an ever increasing part. Whether it is interacting with a video game or simulating a delicate surgical operation, accurate models of real world objects are required. In recent history, models that appear realistic have satisfied the movie and games industries, but for simulations of surgical procedures or fighting forest fires models that behave in a realistic fashion are necessary. Although the advent of the Graphics Processor Unit (GPU) has enabled current models to be realistically rendered in real time, the architecture of these machines is designed around paradigms for modelling that were built decades ago. Paradigms based on meshes of tiny triangles were not designed to represent objects by their volume nor by their behaviour. In this research we seek new paradigms to describe models that exist in a real or artificial world, but can react to their surroundings, behaving realistically alongside real objects in an enhanced or virtual reality situation. Such a paradigm may not fit within the pipeline that is supported by current GPU technology. Indeed, new pipeline architecture could well be a future extension of this research. Much work in computer graphics looks only at solving immediate problems to produce realistic games or movies. In this research we look ahead to future decades where accurate simulation, for example of the human body, and all the physical interactions such a simulation might involve, can be supported in real time. We wish to find new methods to ease the task of representation, design and animation of 3D models, particularly where the model has to conform to a number of physical and aesthetic constraints, and in animation, where the constraints can change in real time. The result will be a reactive model, whose behaviour takes account of the world in which it exists. The near future goal is to extend the concept of implicit modelling, in which models are represented as an iso-surface in a scalar field. Such a field can be generated from geometric and other primitives, which can be derived from data or interactive input, or a combination of the two. By designing suitable functions the combination of fields and their gradients can yield complex, reactive models. Our most recent work combined mesh and implicit modelling to achieve the most accurate, real time, automatic skin deformations to date in an animated character. We wish to extend this work as well as search for a more suitable paradigm that will bring together knowledge from several disciplines, including computer graphics, vision, HCI and artificial intelligence. We also look beyond the theoretical to application areas such as a soft tissue simulation that will enhance training in fields such as neurosurgery.