Nearly everyone of us is surrounded by synthesized virtual environments every day, although it is not always obvious. For example, most graphics in any commercial seen on street posters, on web banners, or in short clips on television contain artificial content and special visual effects to make the communicated message impressive and memorable. In some cases, a certain degree of realism is requested for the virtual environment, especially, if it is mixed with real world content. Present rendering algorithms allow for the production of synthetical images that are hard to distinguish from pictures which were taken from the real world. Even with commodity hardware, high quality renderings can be finished in reasonable time. There are many applications in practice which benefit from virtual environments. For example, many film productions use digital copies of real world locations like famous cities to easily include artificial content (e.g. crowds of people), to allow for complex camera trajectories (e.g. falling from the sky to the street level), or to simulate large scale effects (e.g. a heavy destruction of the scene). This digest of new possibilities would be very expensive or even infeasible without using virtual scenes.

One of the biggest bottlenecks in the production workflow of synthesized virtual environments is the content creation. The modeling of large scenes that provide a high amount of detail is a time consuming and sometimes tedious task. The problem is often compensated by just keeping a high number of artists busy, however, this solution is rather expensive and requires a lot of organizational overhead. Unfortunately, content creation is a creative task and thus, it can not be fully automatized, i.e. the visionary input of artists is still needed.

This thesis is focused on the generation and rendering of cities including common entities like buildings, bridges, and plants. Instead of rationalizing the work of artists, we rather aim at the development of supportive tools that utilize the expressive power of procedural modeling techniques combined with easily operated user interfaces for the intuitive and efficient insertion of creative input. The novelties of this thesis are categorized into three parts. First, the fundamental formalisms of procedural modeling techniques are discussed and analyzed with respect to their application in virtual environments. We introduce an expressive but yet intuitive modeling language that allows for the combination of several modeling strategies. This enables the creation of hierarchical structures (e.g. buildings and plants) as well as of interconnected structures (e.g. bridges and power poles) within one unified description. In the second part, interactive extensions to the previously developed modeling language are presented such that the underlying formalism is not necessarily exposed to the user. Consequently, our approach does not only represent a supportive tool for artists, but even allows users without scripting or modeling experience to combine and manipulate procedurally generated content. In the last part, an efficient rendering pipeline for large scale procedural city models is introduced. Taking a coarse polygonal model of a city as input, high quality facades with real geometric detail are generated on-the-fly such that the massive amount of data never needs to be stored explicitly.

Keywords: Procedural Modeling; L-Systems; Shape Grammars; Interactive Modeling; Realtime Graphics; Screenspace Methods

Selected Topics in Computer Graphics