Research On Ray-casting Volume Rendering Of Large Scale Data Based On CUDA

Volume rendering is one of the prime technologies in the field of Visualization in Scientific Computing. This technology is widely used in medicine, geology, meteorology and so forth because of its advantage in sufficient description and internal details of volume dataset. Volume rendering is significantly meaningful for discovering intrinsic information and analyzing the rules concerning the data. With the progress of data acquisition technology and development of computer hardware, the scale of dataset grows rapidly, GB-level even hundreds-of-GB-level dataset becomes common feature of nowadays volume data, and traditional volume rendering technologies face new challenges. All these facts make us eager for new methods and technologies to solve new problems. Therefore, research on volume rendering has important theory meanings and practical applying value.Research work of this thesis focuses on large scale data, based on CUDA parallel computing platform and takes geological data for example. We discussed several key technologies of large scale data Volume rendering, proposed a ray-casting algorithm of large scale data based on CUDA and developed an interactive rendering prototype system on geological data. On practical request, we also developed stereo display and remote operation&rendering plugins with consideration on the scale of dataset. The main contributions of this thesis include:(?) Organized large scale data into multi-resolution data both on external storage and video memory. Designed a two-stage data manager pattern based on data request and data pre-schedule. Especially for large scale multi-resolution data, we proposed a rendering architecture without presume on data’s structure which makes this architecture adaptive.(?) Designed a resolution-level table which is a data structure based on linear array, and we use this table to present information of3D scene consist of multi-resolution data. This kind of data structure can be easily shared between internal memory and video memory, efficient parallel access is also an advantage. With this structure, we overcome difficulty in sharing traditional tree structure3D scene presenting data between internal memory and video memory, meanwhile, inefficient parallel access on tree structure is eliminated.(?) Taking advantage of CUDA’s high performance, we proposed a real-time ray-casting volume rendering algorithm on large scale multi-resolution data. Compared to traditional ray-casting volume rendering, this algorithm can deal with large scale data and render3D scene consists of multi-resolution data in real-time.(?) Developed an interactive rendering prototype system on large scale geological data, this system demonstrates the generality of our rendering architecture. According to actual demands, we also developed and integrated two functions to our system with consideration on the scale of dataset, the first is stereo display based on3D VISION, and the second is a solution for remote operation&rendering.