New Approach of Vector-based 2D Data Integration

Combining terrain data with secondary data representing man-made and natural features is an essential part in 3D cartography. The easiest way to do this is to drape a map image on a rectangular terrain raster. In this thesis, a new vector-based approach has been proposed that has not been found in literature so far. This approach of creating a combined geometric model from elevation and vector map data is comparatively complex and requires extensive CPU computations. On the other side, the rendering engine is freed from any computations and can be optimized towards dynamically loading and displaying features.

Replacing texture with geometry increases the number of primitives that need to be rendered. There is of course a tradeoff between the number of rendered primitives and the frame rate, which depends on the hardware. At large scales, i.e. for immersive views, superior quality can be achieved this way. Geometry correction techniques such as flattening surfaces have a significant visual effect. Generic textures can be added for simulating the surface mirco-structure of asphalt, grass, sideways, etc. In Virtual Reality simulators, textures are used for modifying material color, reflectiveness, transparency, and surface normals (so-called bump-maps). Also special effects such as ripples on water surfaces are simulated using textures combined with shaders. Although shaders are already supported by 3D exchange formats such as X3D, they are usually not transmitted over networks and compiled in an automatic fashion due to their specific nature, which makes the inclusion of surface effects in interoperable systems quite difficult. Another potential advantage of adding generic textures is to include road markings, which is required by driving simulators and useful for other traffic applications. At smaller scales the scenery gets cluttered because more and more road and land use features need to be displayed. Generalization becomes necessary not only for creating meaningful maps but also for maintaining a constant frame rate. However, there is a trade-off between map quality and degree of generalization, which becomes more important at smaller scales. At a certain scale, map textures provide more information than an integrated terrain at a specific data size. A transition between map textures at small scales and integrated terrain at large scales would be advisable, but has not been experimented with.

Figure 21: Visual comparison of integrated terrain and texture mapping at medium scale (left) and large scale (right),

In general, three different approaches of how to combine vector and terrain data can be distinguished, also pointed out by Bruneton and Neyret (2008):

1. Overlay of geometry on top of terrain (e.g. Wartell et al. 2003). This is often done for displaying GPS tracks, routes, paths, borders and other linear features. However, displaying polygons becomes quite difficult and adds too much overhead.

2. Integration of vector data into the terrain mesh as described here.

3. Rendering features as textures which are then mapped onto the terrain. This can be either done by storing static map images or by consuming vector data and generating texture images on the fly.

Table 1: Comparison of methods for integrating map data with terrain data

Overlay of geometry Integration of vector data Rendering features as textures

Display of micro- structures

Generally not possible Possible by using generic

material textures

Possible to some degree by using fill patterns

Generalization techniques

Using simple Douglas- Peucker Algorithm

Required at medium to small scales using various map generalization and mesh simplification methods.

Often only necessary for small scales using Douglas Peucker and other map generalization methods

Requirements for streaming server

Medium complexity if overlay is done on server side, otherwise only data storage.

Computations relatively complex. Requires large 3D storage system.

Map renderer with medium complexity. Map textures can be stores as

compressed JPEG images.

Adaption to terrain LOD

No Yes Yes

Requirements for vector data updates

Since the terrain geometry remains unchanged, only the overlay must be updated

Updates more expensive since terrain and vector data must be processed

Since the terrain geometry remains unchanged, only map tiles must be updates

Suitable for scene graph streaming

If overlay is done on the server Yes Yes Possible to create semantic models No Yes No Possible to apply custom styles Yes Yes No Method recommended for scales

From large scales at street level to very small scales at global level.

From very large scales (immersive view from pedestrian viewpoint) to medium scale at city level

From medium scale at city level to very small scale at global level

Table 1 tries to summarize different aspects of these methods with respect to the intended usage in a distributed environment involving a streaming server, network transmission, a visualization client, and regular updates.

Which method is most appropriate, depends on the use case, available data and required flexibility of the system. Texture overlay of maps can be implemented quickly and produces satisfying results for most GIS typical applications (Figure 21). Vector-based overlay and integration of GIS data may be preferred for immersive applications and navigation systems. Furthermore, an integrated terrain may be used for semantic modeling. Generated surface partitions can be converted into semantic geometry classes, such as specified by the CityGML modules Transportation and Vegetation.

As a more specific research topic, it was investigated how terrain models can be improved by taking into account the typically flat surface structures of roads and water bodies. The differentiation of land use data sets into feature types can be seen as semantic information, from which typical surface properties can be derived. Large water bodies and parking lots are almost completely flat. Road surfaces may be sloped up to certain degree, which depends on the road type. A unique approach was developed and presented in chapter 4 that modifies integrated terrain

models by applying Finite Element Methods in order to improve the geometrical quality of roads in detailed 3D city models. The vector-based integration of 2D vector data is a precondition for applying advanced mesh manipulation methods. Although this manipulation is not mandatory for enabling 3D Spatial Data Infrastructures, it becomes important for the use case of 3D routing as discussed in section 1.13.2. Results have been significant improvements in the case of the highly detailed model of Heidelberg, producing sharp road edges. In the case of OpenStreetMap, the relatively coarse SRTM base data did not introduce visual errors, that had to be corrected, but the readability of resulting road networks could be improved.