Um framework para a distribuição e renderização de plantas em tempo-real para paisagens em larga escala

Abstract

Rendering vegetation on virtual terrains is an important issue in computer graphics as plants comprehend an integral part of natural landscapes. The main challenges are associated with the complexity of natural ecosystems and the massive number of elements to distribute and render. In particular, plant distribution is often pre-computed, imposing long iteration times during editing and requiring considerable storage space. With the advances of modern GPUs, it is possible to distribute plants at runtime. This dissertation presents a real-time framework to populate virtual scenarios with grass, bushes, and trees. The framework is based on a viewdependent data structure employed to request, manage and render plants dynamically. The inputs of our system are the terrain heightmap and vector features representing natural and human-made distribution constraints such as roads, rivers, and lakes. Based on textures derived automatically from the input data, a GPU-based algorithm evaluates a set of placement rules to determine the plant distributions. The elements generated are dynamically batched and rendered using geometry instancing. The framework allows the quick population of vast terrains with a small runtime overhead.