Investigating ray tracing algorithms and data structures in the context of visibility.

Ray tracing is a popular rendering method with built in visibility determination. However, the computational costs are significant. To reduce them, there has been extensive research leading to innovative data structures and algorithms that optimally utilize both object and image coherence. Investigating these from a visibility determination context without considering further optical effects is the main motivation of the research. Three methods - one structure and two coherent tree traversal algorithms - are discussed. While the structure aims to increase coherence, the algorithms aim to optimise utilization of coherence provided by ray tracing structures (kd-trees, octrees). RBSF trees - Restricted Binary Space Partitioning Trees - build upon the research in ray tracing with kd-trees. A higher degree of freedom for split plane selection increases object coherence implying a reduction in the number of node traversals and triangle intersections for most scenes. Consequently, reduced ray casting times for scenes with predominantly non-axis-aligned triangles is observed. Coherent Rendering is a rendering method that shows improved complexity, but at an absolute performance that is much slower than packet ray tracing. However, since it led to the creation of the Row Tracing' algorithm, it is described briefly. Row Tracing can be considered as an adaptation of Coherent Rendering, scanline rendering or packet ray tracing. One row of the image is considered and its pixels are determined. Similar to Coherent Rendering, an adapted version of Hierarchical Occlusion Maps is used to identify and skip occluded nodes. To maximize utilisation of coherence, the method is extended so that several adjacent rows are traversed through the tree. The two versions of Row Tracing demonstrate excellent performance, exceeding that of packet ray tracing. Further, it is shown that for larger models (2 million+ triangles). Row Tracing and Packet Row Tracing significantly outperform Z-buffer based methods (OpenGL). Row tracing show's scalability over scene sizes leading to a rendering method that has fast rendering times for both large and small models. In addition it has excellent parallelisation properties allowing utilisation of multiple cores with ease. Thus, the Row Tracing and Packet Row Tracing algorithms can be considered as the significant contributions of the Ph.D. These data structures and algorithms demonstrate that ray tracing data structures and adaptations of ray tracing algorithms exhibit excellent potential in a visibility context