使用OpenCL實現多光源方法之全局照明的光跡追蹤

Abstract

本論文之主要目的為在OpenCL平台上實現多光源方法。在圖學領域中，對於如何渲染出擬真畫面的議題一直都有產出相當數量的研究與論文，其中在針對照明方面，可以分成直接照明與全局照明，本研究中所使用的多光源方法即屬於後者，其為全局照明的計算提供了統一的數學框架，在生成虛擬點光源與計算顏色兩步驟皆有可被平行處理的特性，因而建立了我們在OpenCL平台上實行多光源方法的想法。選擇多光源方法除了因為其演算法方便使用者調整虛擬點光源的數量，以達到所需的圖像品質與效能，也因為此方法能適用於各種燈光、材質與複雜模型上，再加上時間成本的可預測性和可靠性，讓多光源方法已成為渲染大規模場景的選擇之一。 本研究針對多光源方法的兩個步驟做平行化，第一個步驟為生成虛擬點光源，即由光源朝向場景中任意方向發射粒子，並透過演算法決定是否在擊中點創建虛擬點光源；第二個步驟為計算顏色，即以像素為平行化單位，透過疊加每個虛擬點光源進行光跡追蹤所計算出的顏色，得到最後的渲染結果。 本研究所渲染的對象為靜態場景，利用OpenCL能夠在異質系統架構中平行且良好地運作的特性，在現今常用的裝置架構下實現多光源方法。我們將傳統多光源方法與光跡追蹤演算法進行改良，實作出具全局照明效果的OpenCL光跡追蹤器。

The main purpose of this paper is to implement many light methods on OpenCL platform. The importance of realistic image rendering with global illumination is underscored by numerous research activities and papers in this field. Many-light methods provide a unified mathematical framework for calculating global illumination. The two-step calculation of this algorithm has characteristics which can be processed in parallel and thus inspires us the idea to implement many-light methods on OpenCL platform. The first step of many-light methods is virtual point light generation, we emit particles in random directions toward the scene from the light source and decide whether the virtual point lights are created or not by the algorithm; the second step is rendering with virtual point lights. By summing up the contribution of virtual point lights, our OpenCL ray tracer gives the final rendering result. The scenes in our research are all static. We use OpenCL platform because it is not only work well at heterogeneous parallel system architecture but also can implement many-light methods on device commonly used nowadays. Our OpenCL ray tracer has improved the traditional many-light methods and ray tracing algorithms which is able to make global illumination rendering.