GPU Volume Raycasting using Bounding Interval Hierarchies

Traditional Direct Volume Raycasting (DVR) on the GPU is limited to uniform voxel grids stored as 3D textures. This approach is not optimal for sparse data sets or data sets with highly varying frequencies because it requires a trade-off between data structure size and the maximum reproducible frequency and it lacks implicit empty space skipping during raycasting. In this paper we present another approach, applying the Bounding Interval Hierarchy (BIH), a hierarchical spatial subdivision of elements traditionally used to accelerate surface raytracing, to volume raycasting on the GPU. Although connectivity information between voxels is lost and the texture filtering power of GPUs cannot be exploited, we show that it may be a viable alternative for DVR and that the approach is generic, allowing all sorts of renderable voxels (not-overlapping finite volume elements/primitives such as cuboids, ellipsoids and truncated Radial Basis Functions) for different tasks like rendering point sets as particle systems (e.g. using spherical voxels) and rendering volumes derived from traditional uniform grids (with implicit empty space skipping and the option for different levels of detail).