運用體素化資訊簡化路徑追蹤之探討

Abstract

想要得到有如照片般逼真的結果，在許多的方法中最常被使用的方法就是路徑追蹤。路徑追蹤是透過模擬真實光線的行進，打出大量的光線進行取樣來得到結果，但是取樣數太少則會出現明顯的雜訊，所以為了得到更加真實的結果，必須花上大量的時間。不過在這之中最花時間的不是直接照明的部分，而是後面進行最終聚集計算的漫間反射，這同時也是造成路徑追蹤耗時的主要原因。同時在許多即時運算的方法中，使用簡化最終聚集的計算來做出漫間反射的效果。 因此本論文藉由OpenCL來使用GPU進行路徑追蹤的加速，而最後聚集使用即時運算的Voxel Cone Tracing來取代，有效的加速路徑追蹤的運算，並保持相當品質的全域照明效果。

To get photo realistic results, path tracing is the most commonly used method in many ways. In path tracing, it sampled a lot of rays to get results by simulating the actual ray. Too few samples cause noticeable noise, so it must be spent a lot of time to get more realistic results. However, the direct lighting doesn't spend most of the time, but the final gathering of the diffuse reflection, which is also the main reason for the time consuming of path tracing. At the same time, many methods of real-time rendering used the simplified final gathering to get the result of diffuse reflection. Therefore, this paper uses GPU to accelerate path tracing by OpenCL, and the final gathering replace by Voxel Cone Tracing. It gets effective acceleration of path tracing, and keep a considerable quality of global illumination.