规则网格实时绘制关键技术研究

发布时间：2018-05-24 08:26

本文选题：规则网格 + 细节层次　；参考：《燕山大学》2013年博士论文

【摘要】：随着现代测绘及遥感技术的发展，获取的高程和影像数据正在以前所未有的速度增长，从数字城市到数字中国再到数字地球，地形数据也呈几何级数增长；与此同时，人类对虚拟场景中几何模型的逼真程度和绘制速度的需求也越来越大。以高程影像和曲面细化模板为代表的规则网格数据具有数据量大、空间布局规整以及需要实时显示等共同特点。目前规则网格实时绘制仍面临诸多技术难点，如缺乏科学有效的多细节层次空间误差阈值设定、缺乏科学的量化精度依据来引导几何数据压缩、瓦块四叉树上视域可见性和细节层次(Level of Detail, LOD)选择计算效率不高、缺少高效灵活的多细节层次网格裂缝处理、曲面细化自适应度低、难以扩展到移动平台等关键问题，制约着数据规模、绘制质量及效率等方面的进一步扩展。考虑LOD空间误差阈值缺乏合理性，在分块地形多误差条件下，以屏幕空间为参考来引导地形层次划分；分析地形处理和绘制过程中引入的误差因素，并以此为参数确定地形数据量计算函数，自适应地设定简化尺度和量化位数，生成最小化的地形压缩数据。针对目前单层次瓦块调度图(Tile Load Map,TLM)存在的缺陷，提出多层次瓦块调度图来加速LOD选择和可见性判断。通过对基于外存地形LOD可视化高效架构特征的分析，基于TLM思想，将瓦块四叉树映射为二维纹理，用GPU完成地形块的视域裁剪和细节层次选择，解决CPU中海量地形地形块选择复杂度太大的问题，形成了多层次瓦块调度图算法；设计∏空间填充曲线重组Tile数据，改善数据访问局部性；设计资源池方式管理纹理对象，减少CPU-GPU带宽。鉴于目前规则网格裂缝处理效率不高，提出多尺度规则网格裂缝参数化处理方法，利用曲线函数控制不同LOD过渡时的网格顶点布局，消除因细节层次过渡产生的T型连接，并扩展至消除多跨度LOD裂缝。分析了二分细化自适应性低的原因，提出在GPU上实现裂缝处理的偶数曲面细化模式。通过在顶点着色器中引入顶点调控函数控制顶点偏移，解决边界裂缝问题；采用参数区域划分子区域的方法，解决函数坐标轴方向不一致引起的浮点误差问题；利用偶数细化层次递增关系，优化参数几何数据布局，达到层次间顶点和索引双重用。给出适用于移动终端的自适应曲面细化管线。在偶数曲面细化模式基础上，，设计适用于移动终端的自适应曲面细化管线，实现了几何模型在移动平台上的交互式渲染。实验表明本文的研究工作有效提高了规则网格的绘制效率，相应研究成果已应用于相关863课题。
[Abstract]:With the development of modern surveying and remote sensing technology, the height and image data obtained are increasing at an unprecedented speed, from digital city to digital China and then to digital earth, topographic data is also increasing exponentially; at the same time, The demand for realistic degree and rendering speed of geometric model in virtual scene is increasing. The regular grid data, represented by elevation image and surface refinement template, have the common characteristics of large amount of data, regular spatial layout and the need of real-time display. At present, the rule grid real-time rendering still faces many technical difficulties, such as lack of scientific and effective multi-level spatial error threshold setting, lack of scientific quantization accuracy basis to guide geometric data compression. The visual visibility and level of Detail, LOD) selection of the quad-tree is inefficient, lacking of efficient and flexible multi-level mesh crack treatment, low adaptive degree of surface refinement, and difficult to be extended to mobile platform, and so on. It restricts the expansion of data scale, rendering quality and efficiency. Considering the lack of rationality of LOD spatial error threshold, taking screen space as a reference to guide terrain level division, the error factors introduced in terrain processing and rendering are analyzed. With this parameter, the calculation function of terrain data is determined, the simplified scale and quantization bit number are set adaptively, and the minimized terrain compression data is generated. Aiming at the defects of Tile Load Map (TLM), a multi-layer tile block scheduling diagram is proposed to speed up the selection of LOD and the judgment of visibility. Based on the analysis of the high efficient architecture features of LOD visualization based on external terrain and the idea of TLM, the quadtree of tile block is mapped to two-dimensional texture, and the view clipping and level of detail selection of terrain block are accomplished by GPU. In order to solve the problem of the complexity of massive terrain block selection in CPU, a multi-level tile block scheduling algorithm is formed; the Tile data are reconstructed by designing the Tile filled curve in the design space, and the locality of data access is improved; the texture object is managed by the design of resource pool. Reduce CPU-GPU bandwidth. In view of the low efficiency of regular mesh crack treatment at present, a parameterized multi-scale regular mesh crack processing method is proposed. The curve function is used to control the grid vertex layout under different LOD transitions, and the T-type connection caused by the transition of detail level is eliminated. And extended to eliminate the multi-span LOD cracks. The reason for the low adaptability of binary thinning is analyzed, and an even surface thinning mode based on GPU is proposed. The problem of boundary crack is solved by introducing vertex control function to control vertex migration in vertex shader, and floating-point error caused by inconsistent direction of function coordinate axis is solved by using the method of dividing sub-region of parameter region. By using even number to refine the hierarchical increment relation, the geometric data layout of parameters is optimized to achieve the dual use of vertex and index between layers. An adaptive surface refinement pipeline for mobile terminals is presented. On the basis of even surface thinning mode, an adaptive surface thinning pipeline is designed for mobile terminal, and the interactive rendering of geometric model on mobile platform is realized. The experimental results show that the research work in this paper has effectively improved the efficiency of regular grid rendering, and the corresponding research results have been applied to the related 863 subjects.
【学位授予单位】：燕山大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：P208;TP391.7

【参考文献】