Volumetric models are used more and more frequently for image synthesis, including photorealistic rendering. Certain highly complex objects, like clouds, can be handled better with volumetric algorithms considering quality and resource requirements. Photorealistic algorithms can highly benefit from these models, since for these, quality is one of the main goals. However classic rendering algorithms cannot deal with such models, therefore different algorithms are needed for these applications. Volumetric rendering algorithms are usually highly demanding; thus, the utilization of the graphical hardware should be considered.
The thesis presents the planning and implementation of an image synthesis system for the GPU, that builds on the Nvidia OptiX ray tracing engine. The algorithms of the system can utilize volumetric rendering algorithms for high quality output. The thesis details different aspects of photorealistic rendering too, like sampling, or camera models by including them in the created models. The volumetric data is handled with the usage of the Nvidia GVDB Voxels framework, and the system is able to generate procedural volumetric models building on the model of GVDB.
The thesis describes the difficulties when dealing with volumetric models – considering representation and usage – through the implemented system. It describes various volumetric rendering techniques and procedural volumetric data generation algorithms too.