A unified volumetric surface representation for 3D imaging and modeling towards efficient integration of geometry and appearance properties

点击翻译按钮获取中文摘要

We are interested in computer vision as an interpretation process of the 3D shape and appearance of objects. From a set of images, the process builds a 3D model, i.e. a geometric representation of objects' surface augmented with its appearance. Similar to the human being who can integrate images in a continuous process, our research aims at developing intelligent vision systems integrating sensors, algorithms and especially, a representation that is suited to the incremental recovery of surface properties. In recent years, we have proposed a new volumetric representation based on vector fields to describe surface geometry. It is proposed to explore the full potential of that representation for integrating these two tightly coupled properties of geometry and appearance.    When geometry cannot be recovered precisely, we explore a new approach that is based on light fields and which adapts the estimated surface geometry within the volume, based on a criterion aiming at improving the quality of the visual simulation of an object. This research lies at the boundaries between computer vision, close-range photogrammetry and computer graphics.    The produced 3D models are used in numerous application fields including industrial design, inspection, rapid prototyping, simulation, heritage documentation, multimedia and arts, biometry and security to name a few. The proposed research will provide the necessary tools for preparing the next information revolution: sharing a virtualized world. Canada is currently among the top leaders in 3D imaging and modeling technologies. Developing a unified modeling approach combining geometry and appearance will reinforce Canada's position.