Interactive Dynamic Environments Using Image-Based Modeling and Rendering

Abstract

This paper presents a method for rendering a dynamic virtual environment using IBMR at interactive rates. Most IBMR methods are designed for use with static scenes and are awkward when extended temporally due to excessive memory requirements or user intervention. In our method, spatial and temporal redundancy are removed from sets of input images to reduce the time and space needed to render an interactive dynamic scene. A pre-processing step is initially applied to remove the redundancy; rendering is performed any time afterward by the explorer. The rendering process is most similar to Layered Depth Images (LDI), while image storage is most similar to video encoding technology. The benefits of our method include the preservation of pixels sampled at higher rates without requiring additional data structures, and the freedom of movement for the user throughout the scene.