Concepts for Automatic Generalization of Virtual 3D Landscape Models

ion. For example, when moving through the model, a LOA management based on the camera distance would need to select high detail geometry in near regions, and coarse geometry in far regions. We found that transparency blending works best when switching between different geometries, compared to other approaches such as morphing. With this technique, generalized virtual 3D landscape models can be created with a LOA that is determined by the underlying hierarchical infrastructure network. T. Glander, M. Trapp and J. Döllner 132 Fig. 4: The comparison shows the original (left) and generalized (right) virtual 3D city model. Landmark buildings are highlighted with their original appearance. 4.3 Generalization Lenses This generalization technique enables the seamless combination of different virtual 3D model variants of the same geographic area (e.g., LOA variants) within a single image using multiple rendering passes (TRAPP ET AL. 2008). It is based on the volumetric clipping technology that enables pixel precise clipping of an arbitrary polygonal scene representation against arbitrary 3D solid clip geometry. This polygonal clip geometry is converted into a layered depth image (LDI) during a preprocessing step. At runtime, the LDI combined with a volumetric depth test is used to determine if a fragment of the rasterized scene geometry is inside or outside the clip geometry represented by the LDI. This test can be implemented efficiently on modern consumer graphics hardware using shader technology. The technique supports simultaneous use of multiple clipping volumes associated with different abstraction levels, can handle overlapping and nested cases, and provides their interactive modification. At runtime, the lens shapes can be scaled, rotated and translated within the scene. The system supports different methods for the lens shape creation. It enables the derivation of complex lens shapes directly from geo-referenced data, such as buffered 2D polygonal shapes and polylines. Further, the lens shapes can also be explicitly using 3D modeling software by importing these through common interchange formats. With respect to the lens interaction, the visualization technique supports scene and camera lenses (Figure 5). While the position of a scene lens is independent from the user’s orientation, i.e., fixed with respect to the virtual environment or attached to a moving object in the scene, a camera lens adapts it position with respect to the current user orientation. It can be used to assure that potential foci are always visible. This minimizes the effort for the user to steer the lenses. Concepts for the Automatic Generalization of Virtual 3D Landscape Models 133 Fig. 5: The examples show three variants of a virtual 3D model at different LOA, integrated in a single image. Lenses can be constructed from paths and regionsof-interest (left) or be camera adaptive (right). 5 Applications of Generalized Landscape Models Possible applications of generalized landscape models include:  Conceptual Landscape Design: It concentrates on early stages of planning processes that require abstract models to express uncertainty and incompleteness as well as the conceptual characteristics of model elements.  Simulation & Analysis: 3D simulation and analysis processes require 3D models as input that have an appropriate and consistent spatial and thematic resolution such as for airflow simulation, waterflooding prediction, hydrologic processes, soil erosion, or vegetative succession.  Mobile Mapping: Generalized landscape models can directly be used in mobile mapping applications and systems to select an appropriate LOA for given camera settings and display resolution. In particular, generalized landscape models can be seamlessly integrated into a single view, blending between LOA variants in a viewdependent/camera-distance dependent way. 6 Conclusions and Future Work The automatic generalization of virtual 3D landscape models involves complex model transformation and presentation algorithms composed of a set of generalization operators that are applied within the visualization pipeline at all stages. The challenges include the automation of the algorithms and their systematic and robust implementation. While many concepts can be taken from cartographic 2D generalization, 3D generalization implies additional constraints and methods to achieve consistent, plausible 3D model variants. The presented concepts have been implemented as research prototypes; we are currently working on an extensible library to provide generalization tools as generic components. T. Glander, M. Trapp and J. Döllner 134 Generalized models aim at providing a consistent and coherent level-of-abstraction among all their components. They facilitate many applications of landscape models such as in conceptual landscape design, in simulation and analysis, and in mobile mapping.