Rendering of natural scenes has interested the scientific community for a long time due to its numerous applications. The targeted goal is to create images that are similar to what a viewer can see in real life with his/her eyes. The main obstacle is complexity: nature scenes from real life contain a huge number of small details that are hard to model, take a lot of time to render and require a huge amount of memory unavailable in current computers. This complexity mainly comes from geometry and lighting. The goal of our research is to overcome this complexity and to achieve real-time rendering of nature scenes while providing visually convincing dynamic global illumination. Our work focuses on grass and trees as they are commonly visible in everyday life. We handle geometry and lighting complexities for grass to render millions of grass blades interactively with dynamic lighting. As for lighting complexity, we address real-time rendering of trees by proposing a lighting model that handles indirect lighting. Our work makes extensive use of the current generation of Graphics Processing Units (GPUs) to meet the real-time requirement and to leave the CPU free to carry out other tasks.

The aim of master thesis is to describe different calculation of global illumination methods including Light Propagation Volumes. All three steps of LPV calculation are widely described: injection, propagation and rendering. It is also proposed several custom extensions improving graphics quality of this method. Two parts of design and implementation are focused on scene description, rendering system, shadow rendering, implementation of LPV method and proposed extensions. As conclusion, measurement and several images of application are presented, followed by comparison in environment with diffenent parameters, thesis summary with evaluation of achieved results and suggestions of further improvements.