Distributed Global Illumination Method Based On Photon Mapping

As the combination of Information Technology and Arts and Humanities, the Digital animation and film industry occupies a large creative industry's market scale among nations. The computer 3D technology is playing an increasingly important role in the creation and production of animation film and television,3D modeling and rendering is becoming a core technology which arouses film and cartoon industries' concerning from all over the world. Speed is a crucial measurable indicator of rendering that has great practical value and research significance for improving lighting effects of games and navigation of big scene.Except using advanced grphic algorithm and optimizing, with the promotion of the computer hardware architecture and maturation of distributed computing, cloud rendering and distributed rendering become new ways to speed up rendering. In other words, to deserialize and ensure distributed rendering by using computing cluster of cloud server and relevant architecture. There are many fruitful research findings in this field; for example, OTOY is an overseas fast and unbiased CPU renderer which is working on a real-time and film-like cloud rendering. It can whip up a movie-like virtual reality closely combined with the OctaneVR. As the leading company in this industry, Autodesk Company's cloud rendering product A360 provide customers chance to draft scene graph on a browser, then submit them to the server to render and get feedback. It's a good method to achieve real-time preview. Currently there are not mature distributed rendering product in China, but the cloud rendering service, a collaboration of Ali cloud and Render Bus, becomes a successful case which combines rendering with distributed computing very well.Render quality is also an important measuring indicator of rendering. A higher real feeling of rendering effect needs advanced rendering algorithm and global illumination method. But these high-quality algorithms like ray tracing are time-consuming processes. Related research fields speed its algorithms up on distributed cluster; they mainly deserialize tasks and use multi-nodes coordinately to finish rendering job. Because of internal characters and serial logicality of light translation, it's difficult to make distributed expansion by using ray tracing algorithm. On the contrary, it will be more realistic to finish expansion work by using a Photo mapping method, because this method has higher degree of Parallelism. In this paper, by using present research achievement for reference, the author selected Mitsuba as renderer platform and made a deep research about how to fulfill real-time photo mapping method on high performance clusters. In the end, practical scheme and the contrast and analysis towards experimental results are achieved.The main aim and content of this paper are:1. Researching the implementation of Path Tracing process of Mitsuba renderer, and make it as the first step of rendering. Judging the attibution of rendering when it meets ray intersection, then deliver tasks to relevant computational node in clusters by socket to complete a rendering task.2. Developing strategies towards rendering task deserialization, including the hardware frame of master and slave nodes for distributed rendering; Data communication protocols; the segmentation, storage and recombination of scene data; the rendering task distribution and consolidation.3. The Photon Mapping global illumination algorithm is proposed as the second step, and chooses two strategies for collecting boundary photon:(1) Make the boundary photon's repeating storage possible by using Overlapped KDTree. (2) Cluster and unite vertex coloring and ask for the photon map of other photons to calculate.By using the strategies mentioned in this article, with a same scene and rendering process and contrasting cluster rendering with standalone rendering, we can get a relative ideal linear speedup and keep the rendering quality identical or nearly identical.