一种基于GPU的遮挡剔除算法改进研究  

A Algorithm Of Occlusion Culling On GPU

作　　者：聂俊岚[1] 郑鹏[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北省秦皇岛066004

出　　处：《微计算机信息》2009年第27期29-31,共3页Control & Automation

基　　金：基金申请人:聂俊岚;基金颁发部门:国防科工委

摘　　要：大规模复杂场景的可见性问题是研究虚拟现实、实时仿真以及三维交互设计等许多重要应用的基础。遮挡剔除算法是可见性问题研究中的支撑技术,也是诸多研究领域面临的一个基本问题。本文在Bittner等人算法基础上,应用八叉树组织场景,通过基于GPU的遮挡查询功能,将多个查询形成一个队列同时查询同时等待结果,减少CPU和GPU的延迟。同时利用遮挡图通过区分被遮挡体之间的相互遮挡关系分别对其进行遮挡查询,提高了查询的效率,进一步减少了绘制面片的数量。实验结果表明,对于大规模高复杂度的场景,该算法确实的提高了场景的绘制速度。The research on visibility for large-scale and complex scenes is under-layer of the virtual reality, real-time simulation and three dimensional interactive designs. The algorithm of occlusion culling not only is the supporting technology of visibility research, but also is the basic issues of many research domains. In this paper, we apply octree structure to organize the scenes base on algorithm of Bittner. In order to reduce the latency between CPU and GPU, we make lots of queries into a query queue and wait the whole result by occlusion query function of GPU. In addition, we process occlusion query for occludees respectively by distinguish the occlusion relation between occludes by occupancy map, so the algorithm reduce more the quantity of fragments. The results show that it improves velocity of drawing for complexity scenes.

关 键 词：大规模复杂场景 八叉树 遮挡剔除 GPU 遮挡图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]