基于外存的场景加速数据结构研究

Research of Out-of-Core based Scene Acceleration Data Structure

摘要:

大规模场景的高效真实感绘制一直以来是图形学绘制研究的难题,特别是当场景规模超过内核内存的容量时. 本文以大规模场景为研究对象，利用GPU的并行计算能力，实现了快速构建多级层次包围盒的并行算法. 该算法针对大规模场景做出了以下贡献，其一利用空间莫顿编码对场景数据进行快速分块并快速构建了一个多级层次包围盒；其二使用分段遍历策略，使用第一阶段的遍历结果进行I/O调度，有效提高了大场景下的遍历效率. 最后通过实验，证明了该算法的正确性与可靠性，并对构建和遍历效率进行了分析.

Abstract:

Efficiently and realistic rendering of large-scale has always been a thorny issue in graphics rendering field, especially when kernel memory can’t hold the entire scene at once. Bounding volume hierarchies is a kind of object-based scene management techniques which is widely used in collision detection, ray tracing system, etc. This paper focuses on large-scale scene and proposes a novel algorithm which named Multi-Level Bounding volume hierarchies, at last implementing it using GPU parallel computing power. Our algorithm has some contributions for large-scale scene. The first uses a linear ordering derived from spatial Morton codes to chip scene data into block extremely quickly and with high parallel scalability; the second is that the traverse efficient have significantly improved by using a two phase traverse tactics, which using the first stage result to control the second stage traverse. Finally, we have done some experiments to proof the algorithm correctness and reliability, also have analyzed the construction and traverse efficiency.