Interactive volume ray tracing

Gerd Marmitt
Computer Graphics Group, Saarland University, Germany, April 2009



Volume rendering is one of the most demanding and interesting topics among scientific visualization. Applications include medical examinations, simulation of physical processes, and visual art. Most of these applications demand interactivity with respect to the viewing and visualization parameters. The ray tracing algorithm, although inherently simulating light interaction with participating media, was always considered too slow. Instead, most researchers followed object-order algorithms better suited for graphics adapters, although such approaches often suffer either from low quality or lack of flexibility. Another alternative is to speed up the ray tracing algorithm to make it competitive for volumetric visualization tasks. Since the advent of modern graphic adapters, research in this area had somehow ceased, although some limitations of GPUs, e.g. limited graphics board memory and tedious programming model, are still a problem. The two methods discussed in this thesis are therefore purely software-based since it is believed that software implementations allow for a far better optimization process before porting algorithms to hardware. The first method is called implicit kd-tree, which is a hierarchical spatial acceleration structure originally developed for iso-surface rendering of regular data sets that now supports semi-transparent rendering, time-dependent data visualization, and is even used in non volume-rendering applications. The second algorithm uses so-called Pl├╝cker coordinates, providing a fast incremental traversal for data sets consisting of tetrahedral or hexahedral primitives. Both algorithms are highly optimized to support interactive rendering of volumetric data sets and are therefore major contributions towards a flexible and interactive volume ray tracing framework.