| Massive terrain data visualization plays a key role in a lot of applications, such as Geographic Information Systems (CIS), 3D map systems, virtual reality systems, training and simulation systems, and 3D games, etc. Typically a visualization system is run in a computing environment consisting of general-purpose computing hardware such as PCs and portable devices (possibly connected via wired or wireless networks). The amount of data the system can render in a fixed period is limited by the capabilities of the hardware. However, massive terrain rendering is very data-intensive since both good interactivity and high visual quality requires large amounts of data. Thus it becomes impractical to process the massive original data set in an ordinary computing environment at rendering time. A simplified data set must be chosen by the visualization system to balance system interactivity and visual quality. Furthermore, the access pattern to massive terrain data is random during visualization since viewers can fly over the terrain surface arbitrarily.; This dissertation addresses the problem of massive terrain visualization from the following perspectives. First, to minimize on-line computational cost for generating view-dependent level-of-detail mesh representation, the concept of per-vertex live range to measure the validity of previous computed results is introduced. Using this live range technique, previous computation results are reused efficiently during continuous view-dependent level-of-detail visualizations. Second, an algorithm based on the visual perception capability of human eyes to simplify mesh representation is presented. The visual metric measures the maximum tolerable luminance change and the maximum tolerable texel deviation in image space. During rendering this metric is used to reduce the number of the rendered geometry primitives by filtering out invisible components. Third, to minimize data out-of-core access cost, a novel clustering algorithm is proposed, which utilizes the level-of-detail (LOD) information for terrain data organization. Last, a simple but effective memory management strategy is introduced to manage data moving from out-of-core to the main memory and from the main memory to the graphics hardware memory. |
