Research Of Acceleration Photon Mapping Based On Gpu

In recent years, with the rapid development of the computer hardware, especially the great progress of the graphic hardware performance, the computer graphics has a considerable development. The improvement of hardware expands the application demand and rendering realism has become a hot topic of people's attention. The photon mapping has become a research hotspot, because global illumination is an important factor in the field of photorealistic rendering and photon mapping is one of the best algorithms about simulating global illumination effect currently.Photon mapping is a simple, flexible and powerful rendering algorithm. But the rendering speed of photon mapping is relatively slow, because it needs large compute during rendering a scene. So it is used only as off-line rendering technique for a long time and it can't use in interactive or real-time rendering areas. With the rapid development of graphics processing unit (GPU) in recent years, GPU already has the efficient parallelism, flexible programmability and efficient general-purpose computing power. So we can take advantage of the power of its parallel computing to accelerate photon mapping rendering now, which makes the photon mapping achieving a real-time rendering performance under certain conditions.This paper makes an analysis and discussion on global illumination and the rendering equation, and then discusses the design process about how to convert photon mapping to a parallel rendering algorithm. Finally, based on the design process, we make a detailed analysis and discussion on the implementation process of photon mapping in GPU and propose some methods to resolve the present problem and difficulty during the implementation process as follows.1. Based on the analysis of the parallel processing method, the paper discusses which processes can be parallel processing in photon mapping algorithm, and how to convert them into parallel operation, leading the algorithm to become a parallel rendering method. The paper also discusses how to distribute scene data on each processor in a reasonable manner, making the processor load as balanced as possible, and speeding up the realization process of photon mapping.2. We make an in-depth study on the programming model of CUDA and analyze the relationship between CPU and GPU in the model of CUDA, and then propose some solution based on the task distribution and load balancing during the process of parallel achieving photon emission and track in CUDA. We also discuss the data structure of photon map and how to parallel storage photon to build photon map in order to help the radiation estimation and scene rendering.3. It needs to use a recursive function to track photons multiply in the realization of photon tracking process, but the present CUDA doesn't support recursive function calls, so the paper changes the recursive thought to recurrence by introducing a queue structure to save and restore the various states during photon tracking, and it resolves the problem of recursive function calls well.4. Based on the detailed analysis and discussion of implementation of photon mapping in CPU and discussion of design method for parallel photon mapping, the paper proposes a method modified from the primitive algorithm, and implement it in GPU completely through CUDA. We make use of the powerful ability of parallel compute to accelerate the implementation of photon mapping. Finally we give out the experiment results and make analysis of them.