Interactive remote exploration of massive cityscapes

We focus on developing a simple and efficient unified level-of-detail structure for networked urban model viewers. At the core of our approach is a revisitation of the BlockMap [CDG*07] data structure, originally introduced for encoding coarse representations of blocks of buildings to be used as direction-independent impostors when rendering far-away city blocks. The contribution of this paper is manifold: we extend the BlockMap representation to support sloped surfaces and input-sensitive sampling of color; we introduce a novel sampling strategy for building accurate BlockMaps; we show that BlockMaps can be used as a versatile and robust way to parameterize the visible surface of a highly complex model; we improve the expressiveness of urban models rendering by integrating an ambient occlusion term in the representation and describe an efficient method for computing it; we illustrate the design and implementation of a urban models streaming and visualization system and demonstrate its efficiency when browsing large city models in a limited bandwidth setting.