Research On Acceleration Methods Of Three-dimensional Display Content Generation Based On Ray Tracing

In recent years,the three dimensional(3D)light field display is a popular auto-stereoscopic 3D display technology.Because the 3D light field display can realize full-parallax and large view angle,it has been applied widely.Multi-view image,which can be realized by the ray tracing technology to ensure the reality of image,is the main display mode in the 3D light field display technology.However,there is a defect for the multi-view image.It takes a lot of time to render a multi-view image due to substantial computation for virtual rays,which is more prominent in super-high resolution display devices,making it difficult to support real-time 3D light field display.To solve the problem,this thesis designs a multi-view image generation method based on dynamic region rendering firstly.This method decreases the amount of computation by updating the dynamic region in the scene to realize real-time rendering in high resolution.Besides,this thesis designs a multi-GPU scheme to accomplish fast multi-view image generation by combining dynamic loading balance and split frame rendering technologies,which can finish fast multi-view image render by allocating computation between GPUs reasonably.In addition,this thesis realizes real-time rendering for multi-view image in double 8K resolution.The main contents and innovations in this thesis are as follows.1.The computation cost of the traditional ray tracing technology is reduced by introducing dynamic region detection.The rendering scene is divided into the static region and dynamic region by analyzing the changing area in it.The computation cost of generating multi-view image can be reduced by detecting dynamic region,and the rendering speed can be improved.For the dynamic region among rendering process,the region in current frame and last frame need to be updated each frame,but for the static regions,it only needs to be rendered in the initial time.The rendering speed can be over 30FPS in 7680×4320 resolution,where the area of dynamic region is less than the rendering scene,and the number of virtual rays is over 30 million.Besides,for different areas of the dynamic region,the proposed method can achieve 4-7 times of rendering speed compared to the traditional method.2.A multi-view image generation method based on path tracing technology and a multi-GPU system is proposed.Path tracing can realize more real effect than ray tracing,but it needs more time to render.By combining the dynamic load balancing with the split frame rendering technologies,the rendering of any frame is performed by multiple GPUs,and each GPU takes almost equal computation.Thus,the rendering speed can be more than 15FPS in 7680 × 4320 ultra-high resolution scene,even though the number of sampling lights is over 100 million.In addition,for different complexities of 3D models,this scheme can achieve nearly twice the rendering speed compared with the traditional schemes.3.Real-time light field image rendering is realized in double 8K ultra-high resolution.For double-8K 3D light field display devices,we use two individual ray-tracing projects to drive them,respectively,and concatenate them together to display.The multi-view images in two display devices are rendered by an individual GPU respectively,and by synchronizing the parameters of the two projects using the UDP protocol,the content of the two display devices can be changed at the same time.The experimental results demonstrate that this method can realize real-time 3D light field display on double 8K display devices while the number of launched lights are 60 million,and the rendering frame rate is more than 20FPS.This thesis provides different solutions to fast render light-field image for the extensive virtual lights over 10 million in three different complex scenarios,and they solve the problem of real-time rendering in ultra-high resolution.It is believed that they provide the novel solutions for high quality 3D light field display.