CyberCity AG, Switzerland
The need for computer-based photo-realistic visualization of 3D urban environments is growing for many applications like urban planning, telecommunication planning, engineering, architecture, 3D city information systems, and homeland security to name only a few. Even though only being able to display 2D data, web portals, where the user can explore digital maps and aerial/satellite images experienced a rapidly increasing popularity.
The development of geospatial solutions has reached a milestone with the upcoming of 3D realtime visualization solutions like Google Earth, Microsoft Virtual Earth and NASA Worldwind. These make it possible for a worldwide community of users to navigate freely in all spatial dimensions from space to street-level in a virtual 3D landscape which consists not only of terrain and image data but also of reality-based 3D city models, integrated into a global context.
3D city models can be cost-efficiently generated using photogrammetry or airborne laser scanner data and – in combination with 3D Digital Terrain Models (DTM) textured with high-resolution orthophotos as complete 3D scene – fulfil specific requirements for above applications and in various industries.
The Swiss/American CyberCity Group is specialized in the production and distribution of reality-based 3D city models. It maintains a constantly growing database of 3D city models of major cities worldwide. Amongst others, these include Los Angeles (USA), San Diego (USA), Chicago (USA), Paris (France), Hamburg (Germany) and Florence (Italy). The modelling of large cities in Asia (e. g. Beijing, Hongkong,…) is planned for 2007 and 2008. However, the experience shows that not only large cities but also small cities benefit from a 3D city model. This article illustrates the efficient generation of reality-based 3D city models and photo-realistic 3D landmarks and shows the real benefit of 3D models in a number of different applications.
Generation of the 3D city models
Creating the geometry
By using photogrammetry it is possible to determine the three-dimensional coordinates of the building roof points by measurement in two aerial images taken from different positions during an flight campaign.
CyberCity-Modeler creates 3D city models using photogrammetry and laser scanner data (LIDAR: LIght Detection And Ranging). Generally, different Levels of Detail (LOD) can be derived from various input data. We differentiate between (see Figure 1) buildings with flat roofs (so called block models), buildings with main roof structures and buildings with detailed superstructures like dormers, chimneys etc. If cadastral data like planimetric building footprints from terrestrial surveying are available, even roof overhangs can be created easily by extruding the 2D polygon to the 3D roof structure. By being flexible in the creation of different LODs from the same data source, it is always possible to tailor the 3D data to be produced to the customer project needs and to its budget range. The accuracy of the final 3D data output depends on the source data. For detailed roof structures it can be approximately 0.20m when using stereo aerial images of a representative fraction scale of 1:5,000 with forward overlap 60% and side overlap 30% or digital imagery with a ground sample distance of approx. 0.10m. On the other hand, if main roof structures are derived from stereo satellite scenes (e.g. Quickbird, IKONOS) with a resolution of approximately 0.60-1.00m, an accuracy of approximately 1.50m can be achieved.
To get photo-realistic virtual building models, different kinds of textures (representative facade picture) may be applied. Large areas can be textured easily using generic textures chosen from an available library of sample regional textures, which brings along the advantage that large areas can be textured with low costs and the disadvantage that specific walls do not appear as they do in reality. To have the real appearance of the actual wall, a procedure to semi-automatically apply facade textures using (oblique) aerial images has been developed. CyberCity applied this texturing procedure for the inner city of Hamburg (Figure 2). This texturing is generally applied for large areas, where the realistic textures are important and a good overall impression has to be achieved, e.g. for web-based 3D geoinformation systems.
Fig. 1 Level-of-Details (A: Block model with flat roof, B: Main roof structures, C: Main roofs with detailed superstructures, D: Roof overhangs).
The most photo-realistic textures are obtained when acquiring terrestrial digital photographs of the buildings on site and the cleaning of the images from disturbing objects such as trees or cars etc. These high quality textures are mostly required when simulating the view of a pedestrian or from a car in simulation and urban planning applications (Figure 3).
Fig. 2 3D City Model of Hamburg, Germany, with semi-automatic texturing using oblique aerial images (Pictometry). Left: St. Michaelis Church. Right: City Hall.'Courtesy of the State Office for Geo-Information and Survey (LGV) of the Free and Hanseatic City of Hamburg'