Efficient implementation of adaptive view-dependent mesh simplification

Abstract

Multi-resolution representation plays leading roles in virtual geographic environment (VGE), interactive simulation and visualization, fly simulation etc. This paper presents an efficient implementation of adaptive view-dependent and topology preserving algorithm with which progressive meshes can be selectively refined or coarsened. In contrast to view-independent algorithms, it allows several different levels of detail to co-exist across different regions of the same object. Regions with great visual importance are represented with high resolution, other parts are represented with lower resolution, and so sizes of LOD models generated with the algorithm presented in this paper are much smaller. In the real-time generation of view-dependent LOD models and its update, parameters such as view point, view frustum, normal of vertices, and position of vertices are all considered. The algorithm mainly consists of two phases: pre-processing and real-time update. In the preprocessing phase, an algorithm based on quadric error metric and edge collapse operator is employed to simplify the original model, and so the original high resolution model can be represented with a coarse base model and a sequence of edge collapse records, which we also call progressive mesh representation. In the real-time update phase, progressive meshes are selectively refined or coarsened according to their visual importance. It is believed that regions fulfill the following two conditions simultaneously are of great importance: (1) intersection of normal cones and view frustum are non-empty; (2) the affecting area of a certain candidate vertex joins current view frustum, and so they will be represented with a high resolution, otherwise, they will be represented with a relative lower resolution. In contrast, the following two characteristics are different from others: (1) Binary trees are stored implicitly in the edge collapse records, so spending on memories is less, most important of all, depth of binary trees constructed in this paper is at least one less than previous algorithms, therefore, traversing of them is faster. (2) Due to the special edge collapse operator employed in the pre-processing phase, precondition for vertex split operations are more relaxed, and so it is easy to get simplified models with fewer amounts of data under the condition of same visual effects.