「地景視覺模擬」可分成地形骨架和地表材質影像兩部分處理。地形骨架一般由規則或不規則的網格構成，主要以地形高程為主，連接地表上特徵點（山峰、山谷、坡度變化較大處等），完整描述高低起伏陡緩特徵；地表紋理則以航空影像或衛星影像捕捉地面紋理物像，搭配敷貼在地形網格上，讓視覺模擬更生動。 本研究利用兩種資料，1m空間解析度的數值地形模型和1m空間解析度的QuickBird全色態影像，分別建立多重細緻程度（Level of Detail, LoD）資料庫，以利視覺模擬時依不同條件動態搭配使用。地形網格部分的LoD，利用四分樹演算法（Quad-tree）找出地形上的特徵點，配合不同門檻值（高差）設定，產生六種不同疏密程度的地形網格；地表材質的部分，則引入影像金字塔概念，以原始影像當作金字塔最底層，也就是最細緻的材質影像，並將資料複製成六份，再縮減其解析度，建立六級不同解析度的資料。 進行視覺模擬時，首先讓使用者決定視覺條件，算出可視範圍。接著採取區塊方式（block-based）拼接適當場景，將前處理產生的網格資料和材質影像，切割成等相同大小的區塊，配合使用者所在的位置與各區塊中心的距離，判斷各區塊應有的資料等級。網格部分是將區塊內的特徵點連成網格，影像部分則拼接不同解析度的區塊影像，最後將兩者結合，完成地景動態視覺模擬。 Terrain representation is one of the most common tasks in visualization of geoinformation. In order to realistically visualize real landscape, several issues have to be addressed, in particular, how to combine terrain meshes and texture images. In this study, Level of Detail (LoD) was applied to both terrain meshes and texture images for producing realistic visualization with better performance but without too much information loss. Several layers of terrain networks and texture images with different levels of details were created from original high resolution digital elevation model and satellite image using quad-tree and image pyramid algorithms, respectively. These were used as base data for dynamic landscape visualization. During the visualization, visible areas of each scene were determined dynamically from viewing parameters. The scene was created using a block-based composition scheme. The study region was divided into several blocks. To improve the visualization performance, only blocks within the view frustum should be processed. In addition, which LoD to use for each block was determined according to pre-defined viewing distance criteria. Then, selected terrain meshes and texture images were combined and rendered to produce the final visualization scene.
