Illumination et ombres pour les forêts pour un rendu temps‐réel

Abstract

The representation of natural scenes and their rendering has always been a challenge for computer graphics, because of their indispensability to many applications such as ecosystems simulators, 3D mapping, flying simulators, video games ..., and because of their geometric complexity, due to very large number of small details they contain which are very difficult to model, and their representation is very expensive in memory, also their rendering requires a lot of computing time. In recent years, graphics hardware has seen a huge development, with the introduction of GPU (Graphical Processing Units), their evolution has been adapted to intensive parallel computing, the development these units aimed to increase programming flexibility, by introducing shaders that are programs executable directly on the GPU, and the ability to locally generate new vertices, which allows enriching the geometric representations of objects in the scene without having to transfer more data between a CPU and GPU, and uses the GPU compute units to do so. This huge development of GPU and these new opportunities may offer the ability for more involvement of geometry in real time rendering plants, if they are properly exploited. To provide a method of plant rendering that maximizes the benefit of using recent GPUs retaining the possibility of using the geometrical representation of the plant on demand, we must first determine a rendering method that is adapted to the rendering very large amounts of geometry by pulling a maximum profit of GPU performances and then define a geometric representation that is best suited to this method of rendering, and then optimize the entire rendering process by exploiting the specific properties of this representation (similarities and preknowledge of the geometry). A very important aspect as the rendering plants in real time is the ability of the application to produce multiple levels of detail, because using the geometrical representation is very expensive, and not suitable for rendering trees that are far from the observer, in this case the application will lose a lot of time rendering details that are not visible to the user, then the render method must be able to produce coarser levels of detail for this case.