high performance computing on graphics processing units: hgpu.org

Metaballs [Bloomenthal 1997] are effective to represent fluids and similar complex and deformable geometries, but their implicit nature makes difficult their visualization in real time. A common strategy is to tessellate the resulting isosurface and to render it on GPU, but it scales poorly as the number of metaballs increases. Kanamori et al. [2008] efficiently raycast thousands of metaballs without intermediate representations. Their method assumes that rays are shot from a single viewpoint, thus preventing secondary effects (no shadows, reflections, etc.), and is limited to polynomial density functions. We propose to exploit the culling capacity of dynamic bounding volume hierarchies (BVH) [Wald 2007], the secant method for ray-surface intersection, and CPU-GPU parallelism to alleviate the restrictions of their method. This results in a general raytracing method, allowing arbitrary ray intersection (visibility, shadow, reflection, refraction, etc.) with metaballs of any finite-support at interactive performances.