| Rendering optimization has always been an important research topic in computer graphic.With the refinement of surface materials and the advancement of global illumination technology,in order to adapt to the requirements of increasingly high rendering performance,in addition to the continuous upgrade of graphics processing hardware,optimization is also particularly important.Occlusion culling is one of the rendering optimization methods,and has maintained a high degree of research interest in the early days.The pre-computed occlusion culling has the characteristics of low CPU cost during the runtime.It can achieve performance improvement at a very low overhead for large-scale scene rendering on mobile platforms,which shows great application significance.First,the scene area is partitioned into cells,and pre-compute the potential visible set of each cell,and do culling according to the potential visible set of the cell where the camera is located during real-time rendering.This article systematizes the culling method,supports an automatic area partition based on digital elevation model and quadtree of scenes,and extended support for multiple rendering methods and custom model filter,and provides system flow control and multiple integrated tools.And by calculating smaller bounding box,using volume to volume sampling method,using low-difference sequences to improve the stochastic process,and parallelization for multi-core processors in the visibility computation,the culling rate and pre-computed performance are improved.In result,a 10-minute pre-computed occlusion culling of the walking perspective in a scene with a camera active area of 640,000 square meters allows he average culling rate to reach 70.7%,requires only 3.7MB of storage space without compression,and the correctness is above 99.94%.With frustum culling,the overall culling rate has even reached 95%.In short,the system cam bring a reliable high optimization under extremely low consumption. |
PDF Full Text Request