GPU-optimized efficient quad-tree based progressive multiresolution model for interactive large scale terrain rendering
Egor Yusov, Вадим Турлапов, Nizhny Novgorod · 2007
This paper presents new effective method for real-time terrain triangulation and visualization. The method utilizes compact progressive terrain representation which is constructed at preprocess stage and requires a very low memory expense. This representation allows easily LOD extraction at runtime. In contrast to previous works our quadtree-based triangulation is not restricted such that neighboring regions must differ by at most one level in LOD hierarchy. This allows constructing more adaptive triangulation. The method performs LOD selection on per-block basis and thus it is not CPU-consuming and fully exploits power of modern graphics hardware. Patch triangulations are constructed only once the first time the patch is needed and cached in fast video memory in optimized form in order to achieve highest rendering performance. The method is driven by the user-defined screen space error and provides guaranteed surface e-approximation. The algorithm supports morphing of both geometry and texture (the last is not usually discussed) which hides popping artifacts related with LOD change. We also discuss some system-level aspects such as multithreaded asynchronous implementation of the proposed algorithm, video-memory allocation control, camera position prediction and speculative data prefetch. The proposed method can be easily extended to allow terrain compression.