Real-time generation of L-system scene models for rendering and interaction

Abstract

We present a framework for generating procedural geometry described by a context-free, parametric L-system completely on the GPU in real-time. These formal grammars allow us to easily describe various types of complex objects, such as vegetation or buildings, in a great diversity. We can create large and complex parts of the scene on-the-fly, which enables us to create a potentially infinite world of such objects. To make modeling easier, we show how the grammar description can be transformed automatically to a shader code that evaluates the L-system on the GPU. Additionally, to allow interaction with the procedural geometry, we propose an algorithm to efficiently perform discrete collision detection with the procedural scene for a large number of objects.