Realistic Vision Rendering: Simulation Of Human Eye Optical System Properties And Depth Of Field Variation

With the rapid development of computer graphics and computer hardware,the rendering result of virtual reality and realistic vision rendering isbecoming more and more realistic. However, the request of the reality of therendering result is also growing. Since human eye is not a perfect opticalsystem, the rendering result for those who have perfect eyes will berecognized by the ordinary people who have refractive errors. Theserendering results will be considered as fake images that are generated bycomputer. In this paper, the human eye optical characteristics includingmyopia, hyperopia and astigmatism as well as the change of depth of fieldthat caused by the brightness of the environment will be simulated to improvethe reality of realistic vision rendering, thus improve the immersion of theuser in the virtual environment.In this paper, the main research and achievements include:(1) An abstracted human eye optical model which contains retina, convexlens and pupil is adopted to simulate the human eye. A novel and effectiveray-tracing-based method is brought out in this paper, which calculates thedirection of rays refracted by the eye without modeling the surface of thecrystalline lens. The parameters of the abstracted human eye optical model arecalculated by the glasses prescription.(2) The requirements that the distribution of sampling points shouldsatisfy are discussed in this paper. In order to achieve better simulation results,a spiral sampling method is proposed, which covers most of the radiationangles and has a distribution similar to Gaussian distribution.(3) The variation of depth of field accompanied by the adjustment of thepupil size in reaction to the luminance of the environment is simulated. Therelationship of the size of pupil and depth of filed is deduce theoretically. (4) The simulation of myopia, hyperopia and astigmatism are achieved byusing the compute shader of DirectX11that compute the ray tracing in theGPU. A parallel algorithm which is suitable for GPU is proposed to sum upthe colors of each pixel in the rendering result and calculate the luminance ofthe environment.(5) Sampling method and the number of sampling points per pixel, whichare two core factors that influence the rendering result, are compared anddiscussed. In addition, the rendering performance is analyzed in this paper,which draws a conclusion that the rendering time is proportional to theamount of traced rays.