Low-discrepancy Sequence Path Tracing

Computer graphics can generate photo-realistic imagery,which involves precise scene model description and realistic rendering.Realistic rendering technology has broad application prospects in film production,image processing and games industrial.The improvement of rendering algorithm can further promote the progress of graphics processing hardware simultaneously.Therefore,computer graphics has a very wide range of scientific and commercial value.The improvement of rendering technology is accompanied by the progress of graphics processing hardware.In the early days,due to the limitation of hardware,rasterization pipeline method was mostly used for graphics rendering.Solving visible surface problem,the ray casting algorithm proposed a method to simulating the light propagation,which is widely used in ray tracing and path tracing algorithms.According to the speed of rendering,rendering technology can be divided into offline rendering and real-time rendering;according to different illumination models,rendering can be divided into global and direct lighting rendering.Monte-Carlo rendering algorithm,which is discussed in the thesis is a very important method in offline rendering.It solves the rendering equation through a large number of random sampling,thus generating photo-realistic computer generate imagery.Mento-Carlo rendering algorithm is the theoretical basis of a several global illumination rendering algorithms,which has high research value.In this thesis,we first study the convergence of Mento-Carlo method.This thesis studies the path tracing algorithm from the perspective of the termination condition.The experimental results show that the termination condition of the algorithm does not significantly increase the number of sampling times,which is instrumental in low-discrepancy sequence path tracing.Based on Mento-Carlo path tracing algorithm,this thesis implements several common illumination models.In rendering,the generation time of low-discrepancy sequence cannot be ignored.The traditional generation method of low-discrepancy sequences based on inverse racial function has the disadvantage of too long generation time.Moreover,the stratification effect of the Halton sequence on the high dimension has a negative impact on sampling.This thesis also proposes a new low-discrepancy sequence generation method based on Halton Sequence,which reduces 75%time of generating a low-discrepancy sequence than radical inverse function method.The new sequence distributes more evenly in space and reduces the stratification in high-dimensional case.Based on the new low-discrepancy sequence,a imporved path tracing algorithm is proposed.In statistics,the distribution of new low-discrepancy sequence is consistent with random sequence,so the image generated by these sequence can converge to the same benchmark image.According to experiment result,this method does not increase the number of sampling times;Furthermore,deterministic sequence is more easily applied to parallel implementation.Another usual method to speed up the rendering process is to use parallel computing.According to the path tracing,each ray tracked is independent.Parallel computing can significantly accelerate the rendering process.Based on this,finally,this thesis implements parallel path tracing algorithm based on GPU and CPU.Deterministic low-discrepancy sequence can easily implement path tracing in parallel computing.The difference of CPU and GPU structure leads to the difference of acceleration effect between the two kinds of hardware in parallel rendering.According to experiment result,parallel computing greatly improves the rendering speed and parallel path tracing has a better performance in GPU rendering based on CUDA.