Exploiting Coherence in Voxelized Geometry

Abstract: In computer graphics, the geometry of virtual worlds can be represented in numerous ways, from collections of simple triangles or voxels to higher-order primitives like curved surfaces. There is a trade-off, both in terms of memory consumption and processing time, between the cost of an individual primitive and the number of primitives required to faithfully represent the world. This thesis focuses on decreasing the memory consumption of large collections of very simple voxels, while still maintaining high processing performance. The thesis includes three papers concerning voxel data of static geometry, shadow volumes, and time-varying geometry, respectively. A central idea in all the papers is the utilization of coherence in the data to reduce the overall size. Coherence is utilized when there are many identical subvolumes. The identical subvolumes are identified without manual intervention and only a single instance of these volumes is kept, letting several parents reference a single unique subvolume. The voxel data is represented with hierarchical information in a directed acyclic graph, which allows for fast traversal in, e.g., a ray tracing application. The directed acyclic graph structure is further optimized for high performance of many localized look-ups (like a PCF-kernel for hard shadows) and for even lower memory footprint when streaming time-varying geometry.