GPU-Based Real-Time Procedural Distribution of Vegetation on Large-Scale Virtual Terrains

We present a novel framework for real-time procedural distribution of vegetation, capable of handling large-scale terrains. Our approach considers several natural aspects that influence the adaptability of each plant type to topographic and environmental factors displayed across the terrain, as well as interactions between different plant types. The adaptability of each plant type is modeled through a set of consistent parameters that afford full control to the user over the final results of the distribution process. The proposed architecture relies on GPU parallelization and GPU instancing to improve performance. Our framework can be used to generate the vegetation cover of a terrain at runtime or to create an initial distribution that could latter be manually edited, expediting the process of decorating large environments. The results show that our framework can achieve natural looking vegetation distributions, while maintaining the computational costs compatible with realtime applications.