Graphics technology has always been limited with the rendering performance of virtual worlds. Performance is dependent not only on the hardware capabilities, but also on the specification of non-visible objects. The simplest solution requires detecting visibility for each object individually, which can be very time consuming, especially in large virtual worlds. To overcome these problems, new kind of algorithms were developed, which are called scene graphs. The most common trait of this group of algorithms is space subdivision into well defined regions. With this process the question of object visibility is transformed into region visibility, which can be a lot more efficient. The aim of the thesis is to compare various scene graphs in different worlds that occur in modern video games. We showed that each of the scene graphs used increased performance drastically in each world, but there is a lot of differences between them. We also observed that the most important factors influencing scene graph performance are camera position, direction and world composition.
|
