Representation techniques for real-time visualization of hybrid terrain models

Resumo

In recent years, advances in sensing and data processing technologies have led to a continuous growing of geographic datasets. Interest about using this increasingly detailed geographic information, has consequently arisen in application domains such as urban planning, cartography, global navigation systems or virtual reality. However, real-world digital terrain models demand for efficient techniques for managing and rendering their complex and large-scale geometric data while, at the same time, may require specific features and operations. An effective terrain model should therefore maintain a good trade-off between the accuracy with which the terrain surface is modeled, storage requirements, and real-time rendering. Multiresolution representations try to solve the problem by reducing the rendering cost without a significant loss in the visual appearance of the scene. Unnecessary detailed data is reduced by using alternative representations of the same model with different accuracy, simplifying the polygonal geometry of distant or unimportant portions of the model. A different strategy, barely studied until recently, is to use hybrid terrain models combining large regular data sets for large unimportant areas of the terrain, and high-resolution irregular meshes for topographically complex terrain features and man-made microstructures. A more precise geometry description of morphologically important terrain features is thus provided, enhancing the perceptual quality of visualized terrains models without increasing the overall resolution of the whole mesh. In this thesis several techniques to generate and visualize 3D hybrid terrain models are proposed. These techniques can integrate geographic data from heterogeneous sources, combining the benefits of coarse-grained grid-data and fine-grained irregular data without requiring a remeshing process that would modify the original datasets. Multiresolution rendering of the obtained representations are supported in the large regular areas, and in some cases in the whole hybrid model. All these methods are based in a local tessellation approach which can selectively rebuild the adaptive tessellation between the regular and irregular meshes of the models during the interactive visualization process. This approach was originally developed in the Hybrid Meshing algorithm developed by some members of the research group, which is the departing point of this work. In particular, this thesis contains an analysis of the Hybrid Meshing algorithm and two different proposals of an interactive hybrid terrain model renderer based on it. Since the original algorithm proposes a theoretical hardware unit in the graphics pipeline to assist in the rendering of the hybrid terrain model, the algorithm have been adapted to a conventional graphics pipeline in two different ways. One of the methods is complete faithful to the Hybrid Meshing algorithm, while the other avoids the preprocessing phase by using a conventional tessellation algorithm. The Hybrid Meshing algorithm has been also enhanced and adapted to run in a parallel implementation using the Geometry Shader Unit of the GPU. This parallel proposal incorporates several improvements in the organization of the data lists and in the interactive visualization phase. Besides the several extensions and improvements to the Hybrid Meshing algorithm, two new methods have been developed to overcome the limitations of the previous approaches: the EHM and the EDP proposals. The EHM algorithm supports multiresolution rendering of the whole hybrid model, both in the base regular grid as well as in the high resolution irregular meshes. Therefore, the EHM algorithm is the first method in the literature achieving a complete view-dependent rendering of a hybrid model without cracks or other visible artifacts in the visualization of the model. The EDP algorithm proposes the use of the external edges of the irregular meshes as the fundamental primitive to tessellate the boundaries of both types of meshes. This approach achieves a better parallelism in the rendering phase while simplifies the preprocessing phase by minimizing the insertion of vertices in the irregular mesh.