The recent large availability of geometric and photgraphic data of urban models,
coupled with an increased user expectation, lead the computer graphics research community to focus on novel methodologies for real-time visualization of massive citiscapes. Such 3D datasets are particularly hard to deal with due to their relatively simplegeometric structure (thousands of buildings, each represented by few simple polygons, for which common simplification techniques do not apply) and a huge image data footprint (aerial and street-view photographs).
We present a novel technique, the BlockMap, aimed at efficiently encoding and rendering urban models in a multiresolution framework, exploiting hardware accelerated ray-casting. The BlockMap data structure encapsulates all relevant information for reproducing typical urban scenarios, such as vertical façades, flat and sloped rooftops and the terrain surface. It also encodes shading information by storing surface normal and color, and indirect global illumination terms such as ambient occlusion. We employ an image based preprocessing methodology which allow us to create a multiresolution representation of the entire city model. Being the data preprocessed and rendered in an out-of-core fashion, our technique has been proven to allow for interactive fruition of very large dataset, even across network infrastructures (streaming).