ADVANCED GRAPHICS TECH & QUALITY OF EXPERIENCE FACTORS IN VIRTUAL ENVIRONMENTS

对于漫游来说，许多模型太复杂了，无法进行交互， 这需要至少20帧/秒的更新速率。如果，对于 例如，其中一个图形引擎能够渲染200万 多边形数/秒，超过10万个多边形数的模型不能 仅靠蛮力就能平稳地穿越。在1995-96年度，我们 开发了用于制作交互式漫游的各种技术 大型模型的可能性。在1996-97年间，我们研究了如何 将这些技术中的几种集成到单个渲染中 框架。这不是微不足道的，因为一般而言，每种技术 需要其自己的特定数据结构；这些不兼容；以及 机器内存不能包含海量数据的多个数据结构 模特。我们取得了相当大的成功。8月，在一年一度的 SIGGRAPH会议，我们演示了交互式漫游，速度很快 4-20帧/秒，一个由大约1300万个多边形组成的巨大模型。 使用的技术之一是在模型空间中填充一些 10,000个单元格沿着大多数观众想要走的路径排列。 对于每个细胞，构建了一个盒子，每侧细胞大小约为6倍。 长方体外的所有几何体都渲染到关联的网格上 对于每个盒子的墙壁，使用投影纹理作为前处理。在… 运行时，只为所有对象渲染简化的投影纹理 查看器当前所在的长方体之外的几何体。这个 长方体内的几何体使用细节级别进行渲染 简化，如上文子项目9所述。仅几何体 在视图中渲染平截体；分层剔除技术 使这些决定能够非常迅速地做出。又一次 分层剔除技术用于避免呈现 被更靠近的单个对象或对象组遮挡 观众的眼睛。

Many models are too complex for walkthroughs to be interactive, which requires an update rate of at least 20 frames/second. If, for example, one has a graphics engine capable of rendering 2 million polygons/second, models of more than 100 thousand polygons cannot be smoothly traversed by brute force alone. Over the 1995-96 year, we developed a variety of techniques for making interactive walkthroughs of massive models possible. During 1996-97, we studied how to integrate several of these techniques into a single rendering framework. This is non-trivial, because in general each technique needs its own particular data structure; these are incompatible; and machine memories cannot contain multiple data structures of a massive model. We achieved considerable success. In August, at the annual SIGGRAPH meeting, we demonstrated interactive walkthrough, at speeds of 4-20 frames/second, of a massive model of some 13 million polygons. One of the techniques used was to fill the model space with some 10,000 cells that lay along the paths most viewers would want to take. For each cell, a box, some 6x cell size on each side, was constructed. All the geometry outside the box is rendered onto meshes associated with each box wall, using projective textures, as a pre-process. At run time, only the simplified projective textures are rendered for all the geometry outside the box the viewer currently inhabits. The geometry inside the box is rendered using level-of-detail simplfication as described in Subproject 9 above. Only the geometry in the view frustum is rendered; a hierarchical culling technique enables these decisions to be made very rapidly. Yet another hierarchical culling technique is used to avoid rendering objects that are occluded by single objects or groups of objects nearer the viewer's eye.