在行動裝置上達成複雜場景之互動性基於物理高品質成像

Abstract

過去因為手機在計算速度與記憶體容量上的限制，遊戲中為了達到即時互動的效果，大多只是將場景與人物以貼圖的方式顯示在畫面上，或是使用光柵化（Rasterization）的方法產生陰影與直接光照的效果，但很難有反射、折射、散射等讓玩家有臨場感的逼真畫面。 如今電腦上實時光線追蹤技術已趨於成熟，手機將不可避免地成為下一個目標。在此篇論文中，我們使用 OpenGL Compute Shader，利用手機 GPU 的平行計算能力來實現路徑追蹤（Path Tracing）；此外我們編寫了加速結構（BVH Tree with SAH 、Huffman Tree）來減少找尋三角片交叉的耗時，並使用「無堆疊」的方法作遍歷，因此可以在SIMT架構的GPU上高效運行，更實驗了不同的光線追蹤方法（Bidirectional Path Tracing、Metropolis Light Transport）來加速收斂場景中的顏色；如最後的成果所呈現，即使是擁有豐富材質的複雜場景，我們的程式依舊可以在很短的時間內畫出雜訊少又逼真的圖。In the past, due to the limitations of mobile phones in terms of computing speed and memory capacity, in order to achieve real-time interaction in games, most people showed textures on the screen, or generated effects of shadow and direct lighting by rasterization. But it is difficult to have realistic images such as reflection, refraction, and scattering that give players a sense of presence. Now that the real-time ray tracing technology on computers has matured, mobile phones will inevitably become the next target. In this paper, we use OpenGL Compute Shader to achieve path tracing on the mobile phone. We implement Bounding Volume Hierarchy acceleration structures with Surface Area Heuristic and Huffman Coding to reduce the search time for triangle intersection. We use Stack-less BVH Traversal to run on the SIMT architecture efficiently. In addition, we experiment with different ray tracing methods such as Bidirectional Path Tracing (BDPT), and Metropolis Light Transport (MLT) to accelerate the convergence of image colors. The final results show that even for complex scenes, our program can still draw a realistic and less noisy picture in short time.