Parallel Direct Volume Rendering of Intersecting Curvilinear and Unstructured Grids Using a Scan-line Algorithm and K-D Trees


Scientific visualization of volume data is increasingly being used in applications such as Medical Imaging and Computational Fluid Dynamics (CFD). Due to the irregular nature of the geometry of CFD applications, volume data for CFD applications is often stored in curvilinear or unstructured grids. This increases the computational and memory resources needed to render the volume due to the limited amount of geometrical coherence that can be exploited. A method of rendering multiple, possibly intersecting, curvilinear and unstructured grids is presented. This method, which has been implemented for a MIMD parallel architecture is based on a scan-line algorithm and provides ray-cast quality images efficiently by taking advantage of polygonal coherence between scan-lines. It is capable of rendering intersecting grids without the computational and spatial expense of subdividing intersecting cells, which most ray-casting and all cell projection methods must do. Further efficiency is obtained by the use of a k-d tree with prefix-order partitioning of triangles, to reduce the number of primitives that must be processed for one rendering. All rendering is done in software which eliminates the need for specialized graphics hardware. The performance and rendering quality of this algorithm are compared with hardware- assisted cell projection algorithms, for both rectilinear and curvilinear grids, and with a ray-casting algorithm for rectilinear grids.

My thesis (1.0 Mb pdf file)