Displacement Mapped Billboard Clouds

This paper introduces displacement mapped billboard clouds (DMBBC), a new image-based rendering primitive for the fast display of geometrically complex objects at medium to far distances. The representation is based on the well-known billboard cloud (BBC) technique, which represents an object as several textured rectangles in order to dramatically reduce its geometric complexity. Our new method uses boxes instead of rectangles, each box representing a volumetric part of the model. Rendering the contents of a box is done entirely on the GPU using ray casting. DMBBCs will often obviate the need to switch to full geometry for closer distances, which is especially helpful for scenes that are densely populated with complex objects, e.g. for vegetation scenes. We show several ways to store the volumetric information, with different tradeoffs between memory requirements and image quality. In addition we discuss techniques to accelerate the ray casting algorithm, and a way for smoothly switching between DMBBCs for medium distances and BBCs for far distances.