In-Core Level-Of-Detail Generation For Point Clouds On GPUs Using CUDA

Rendering arbitarily large point clouds requires level of detail (LOD) structures that allow users to stream and render representative subsets of the data in real-time. This thesis introduces PotreeConverterGpu, a GPU-based software for generating octree-based LOD structures from point clouds. It is based on PotreeConverter and investigates GPU-friendly algorithms to generate the same structure up to 85 times faster for similar quality. The thesis also explores color-filtering for lower levels of detail to improve the quality for point clouds in a fashion similar to mip mapping for textured meshes.