Ray tracing acceleration structure : GPU based KD-tree using SAH heuristic

• Main Author: Nguyen Khac Trung, Nguyen.
• Other Authors: Seah Hock Soon
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Computer graphics
• Online Access: http://hdl.handle.net/10356/52063
• Institution: Nanyang Technological University

Ray tracing is a very widely used rendering technique in many applications in the movie industry, architecture and media industry. The reason why ray tracing is so popular is because it can produce the most realistic and most accurate images compared to other rendering techniques due to its straight forward nature of emulating light. However computational costs prevent it from being widely used in interactive applications. Much research has been done to help accelerate the ray tracing process and one of the most popular data structures being used is the KD-Tree. However KD-Tree construction is also a computationally expensive task thus holding it back from being used in interactive applications in which the scenes contain animations and require the tree to be rebuilt in every frame. With the advance of multi-purpose graphics hardware with its massive parallelism it is now possible to accelerate the ray tracing process further to improve the performance greatly from traditional CPU based approaches. The GPU construction stage also benefits greatly from the computing power of modern graphics hardware and thus it is now possible to construct KD-Tree in real time. In this final year project we try to implement such an approach by making a n early full GPU based ray tracer using KD-Tree as the acceleration structure. The whole KD-Tree construction algorithm is also implemented to try to take advantage of the GPU as much as possible using the CUDA framework running on NVIDIA graphics cards. Due to time constraints our GPU ray tracer is quite simple and does not support animation, however the code is functional and reasonably stable. Further optimizations can be made to improve the performance and more features can be implemented in future projects.