Parallel Path Tracing Algorithm Based On Adaptive Sampling

In recent years,with the development of computer hardware and graphics,the film and television animation industry has been greatly developed.At the same time,3D technology plays an increasingly important role in the process of film and television animation production.The highly realistic rendering technology is one of the core links of 3D film and television animation production.The purpose of highly realistic rendering is to generate real photo level images for CG scenes.As the computing power increased,researchers begin to focus on accurately simulating light propagation to create images that are indistinguishable from real scenes.The geometric optical model that calculates all the propagation paths of light energy in the scene is called the global illumination model.Path tracing is one of the basic methods for achieving global illumination rendering,whose goal is to transform the input 3D scene into an image with lifelike light effects.Path tracing can simulate the physical effects of reflection,refraction,and scattering caused by the interaction of the light with the object in the scene by reversing the path of the light propagating in the real world(the path from the viewpoint to the illuminator).Path tracing is a research hotspot in computer graphics.The realistic rendering based on path tracing has a wide range of applications and research values in virtual reality,film and television special effects,3D simulation and other fields.Currently,path tracing use uniform ray sampling widely,with each pixel sampling the same amount of light.This method has the contradiction between image quality and calculation time:low sampling rate,short calculation time,image noise and poor image quality;high sampling rate,low image noise,high image quality,but long calculation time.With the development of 3D games and virtual reality,the interactive image-based real-time application requires higher and higher image quality.This requires that the number of light emitted by the optical path tracing pixel reaches K level,the amount of illumination calculation is large,and the drawing time is long.It is urgent to study the parallel optical path tracing method to accelerate the drawing process,which plays an important role in accelerating the production process of the film and television animation industry.At present,most feasible parallel methods are to divide the imaging area into blocks,and to calculate the task primitives in parallel,and to divide the computing tasks into multiple nodes for parallel computing can effectively reduce the rendering time.This mode is convenient to implement,but the parallel granularity is coarse,and the calculation time of different rays is not the same,resulting in a larger calculation time difference between nodes,and the load unbalancing between the nodes.To improve the noise problem in uniform sampling and the unbalanced load problem in parallel tasks,a novel parallel path tracing algorithm based on adaptive sampling is presented in this paper.At the stage of pre-processing,the frequency domain information of the ray is obtained based on the covariance tracking method and then the corresponding adaptive sampling count of each pixel is acquired.At the same time,the estimated complexity of a single ray is obtained based on the ray frequency and the deep of path tracing.At the stage of processing,calculating the complexity of the ray tasks for each block region on the drawn image by using the estimated data at the stage of pre-process,and then the tasks are distributed according to the complexity.In this way,the total task complexity obtained by each node is almost equal to each other,ending up with load balancing.Experimental results show that,compared with the conventional uniform sampling method in path tracing,this method can not only reduce the drawing noise under the premise of emitting the same amount of rays,improve the rendering quality but also ensure the load balancing among the nodes in parallel drawing.It is highly scalable and can achieve an efficient speedup.