Analyzing performance for lighting of tessellated grass using LOD

BackgroundThere are different methods for rendering grass in real-time, for video games. This thesis looks at an algorithm that uses hardware tessellation to generate geometry during run time. These concepts are explained in more detail in the introduction chapter as well as the method chapter. The objectives problem that this thesis revolves around is this; When rendering grass using tessellation a lot of geometry is created. This means that the grass might be a bottleneck, meaning it eats too much performance. This applies especially when adding lighting to a scene dense with grass. Lighting requires a lot more computations, so adding this can create a difference, a "dip" in performance. To try and aid this dip, LOD in different forms will be added. Then tests will be carried out to see if the divide is lessened. This would allow for a lot of lighting to be used together with this expensive technique. LOD means a way to lessen the geometry being rendered through different algorithms so that the performance cost will not be as high. methods answer the relevant research questions, a literature review was carried out. This is to understand the relevant 3D techniques better. And also to learn more about rendering grass through tessellation. Then an implementation was conducted, with this experiment to test different LOD techniques with and without lighting. Each scenario was tested ten times and the average framerate was measured and put into charts.ResultsThe average framerate dropped when light sources were added in all of the tests. The binary tree consumed more RAM usage but aided the framerate more than the quadtree. The VRAM usage for the grass tessellation was around 100 MB.ConclusionsFor future work, it would be interesting to compare different ways of rendering grass for the sake of user value. Do people prefer tessellated grass when minimal lighting is applied? Or do people prefer quads to represent grass? The results did not showthat the LOD implemented solved the issue presented in this thesis