Large-scale Terrain Rendering Method With GPU-based Dynamic Error Metric

Terrain rendering is a subject of displaying, simplifying and simulating of dataterrain model (DTM) or data elevation model (DEM), which is a branch of computergraphics. Terrain rendering is normally based on DEM data, dynamically displays3-dimensional approximation of terrain, and widely used in the field of flightsimulation, battlefield simulation, three-dimensional games and scientificvisualization, etc. After decades of development of terrain rendering technique,resulting in a number of excellent algorithms, but with the rapid development ofgraphics hardware in rendering speed and programming performance, how to take fulladvantage of the hardware to further enhance the effect of terrain rendering, hasbecome the topics that must be faced by the field of terrain rendering.In recent years, some terrain rendering algorithms had begun to take advantageof GPU patch rendering capabilities, to improve the rendering performance. But mostof these algorithms only use the fast rendering capability of graphics hardware, notuse programmability of the GPU. And error metric is too simple to generatedoptimized terrain model. Although these algorithms can achieve effects and frame ratefor large-scale terrain rendering, but with the further increase in size and resolution ofterrain data, the drawing effect is difficult to guarantee.In this paper, we took the key issue as the study objective, based on intensivesurvey and comprehensive analyses of various approaches to terrain rendering, inaccordance with the improvement of graphic hardware in rendering performance andprogrammability, and proposed a dynamic large-scale terrain rendering method withGPU-based error metric to increase resource utilization of Graphics hardware andenhance rendering effect. We mainly study about the GPU-based dynamic error metricto construct adaptive terrain model, GPU-based triangulation strip construction,incremental data transmission, etc. The main features of our method are as follows: 1. GPU-based dynamic error metric and terrain representation.Based on the concept of the restricted quadtree and saturation errors for regulargrid terrain data, first of all, the terrain data is divided into tiles of equal size. Inpreprocessing, based on error saturation concept, we construct the nested geometricerrors and nested bounding sphere radius for each tile. During run time, according tothe parameters of the point of view and the GPU vertex-based error metrics which isbuilt from the concept of nested spherical inherited, the top-down select of the meshvertices is used to build the terrain model. And at the same time, formming thetriangle strip to draw on screen. This approach supports view-dependent dynamicerror metric calculation. By taking advantage of existing GPU programmingtechniques, both error metrics and triangulation strip are processed by GPU toimprove efficiency of large-scale dynamic terrain rendering algorithm.2.GPU-based dynamic triangulation strip construction.For every tiles, a look-up table (LUT) is built by special sequence of terrainvertex. At run time, the terrain vertices which are selected by dynamic error metric,compare with LUT and directly generate triangle strip sequence. The operationsupports parallel computing requirements and can be processed on GPU too.3.Incremental data transmission.In our method, terrain data is sorted according to the saturate error, which iscalculated during preprocessing. At run time, a simple pre-judging method is executedto judge transmitting amount of terrain data. When view point is changed, terrain datacan be transmitted incrementally. Although the pre-judging compution is processed byCPU, but the time cost is very little. And Incremental data transmission can greatlyreduce the amount of real-time transmission of data to cut down the demand fortransmission bandwidthIn summary, the research results of this article provides a new way for terrainrendering method on the basis of modern graphics hardware, which has a certaintheoretical significance and application value, and provides a useful method and technique for real-time terrain rendering.