Paul Anderson and Choi Byung-Geun

Two papers which describe the research currently going on in the area of procedural architecture are “Populating Ancient Pompeii with Crowds of Virtual Romans” (Maïm et al.) and “Rome Reborn 2.0: A Case Study of Virtual City Reconstruction Using Procedural Modeling Techniques” (Dylla et al.).

Maïm et al. were primarily interested in populating their model of the ancient city of Pompeii with people to show what everyday life would have looked at. This is a little bit different from other applications of procedural architecture, but the people generated are still following a set of basic rules, so it is generally the same process. The rules include things such as looking in shop windows and emerging from shops with bread or a jar of oil. The people also had set destinations and were routed to them using the shortest path along the city streets. One problem faced by the authors of this paper was the limits of computing power. It was too taxing to render all of the buildings and people at once, so the computer kept track of but did not render the positions of buildings and people out of the users field of vision. Things in the users field of vision were rendered at different levels of detail (deformable meshes, rigid meshes, and imposters in declining order of detail), and the level of detail depended on the distance from the users viewpoint.

Dylla et al. talked about their research group’s model of ancient Rome and the improvements they gained in moving from a hand-modeled version 1 to version 2 which incorporated procedural architecture to fill in buildings in areas without historical documentation. Version 2 of the model still had a distinction between Class 1 buildings (landmarks such as the Forum and the Coliseum) and Class 2 buildings (where no architectural drawings are available). The type of Class 2 buildings (single-family house, apartment, warehouse) was known from the “regionary catalogues” of Rome. This helps the procedural architecture process to be more accurate, since we know the building type is correct so the building generation is not totally random. With respect to the first version, the procedural architecture method was much faster than modeling the class 2 buildings by hand, and the end result was faster to model as well. The rendering for the model was done on a server, which is much more powerful than most users’ machines. This allows for a more complex and detailed model, since the rendering is done remotely and the individual users only need to access the part of the model in which they are currently located.