Efficient GPU usage for rendering of Volume Data

Abstract

Visualising medical images or scientific data that can be shown in 3D poses some interesting challenges. This thesis investigates two distinct areas of Direct Volume Rendering (DVR); time-varying datasets on emerging light-field displays, and acceleration data-structures for large, static datasets. Firstly, a novel method to efficiently display volume data that changes over time (time-varying) on emerging display technologies is shown to improve upon traditional techniques in the field. The second main focus is acceleration data-structures in the context of volume rendering. The attention is on how particular acceleration data-structures are used in current DVR approaches and why Bounding Volume Hierarchies (BVHs) haven t been the method of choice for large-scale volume rendering. This thesis provides an insightful look at how new hardware implementations in consumer products impact on DVR, and how this may change the adoption of BVHs in this area. Results show that BVHs coupled with a pre-clustering step are a viable alternative to standard octree Empty-Space-Skipping (ESS) techniques.