ABSTRACT: Many communities throughout the world have captured their infrastructure in GIS databases, yet they’re faced with another challenge: visualizing the complex 3D urban environment, with its myriad of structures and interconnections from 2D data layers. A varied user community, consisting of city planners, commercial developers, emergency first-responders, utility companies, and A&E firms, increasingly requires support for a 3D interactive presentation of the urban environment. TerraSim, a Geospatial IT company, has been working on applications within its TerraTools® product to provide that support for several years. This article discusses some of the considerations involved in constructing a 3D virtual environment.

Reconstructing the world

While many communities throughout the world have now captured their infrastructure in GIS databases, they are faced with another challenge: visualizing the complex 3D urban environment, with its myriad structures and interconnections, from 2D data layers which uniformly represent objects as abstract colored points or lines. Experts learn to “reconstruct the world” mentally, but this reconstruction cannot be communicated quickly and effectively to decision makers.

The solution is to depict the city in 3D from its GIS representation. However, important visual and geometric information is necessarily discarded in collecting the GIS data. To construct a realistic representation of the city we must reverse the abstraction process which produced the GIS, fully leveraging GIS information while reconstructing the 3D world in a principled manner. In this way we can produce a visualization faithful to reality and representative of the physical world as we navigate through it.

A varied user community, consisting of city planners, commercial developers, emergency first-responders, utility companies, and A&E firms, increasingly requires support for interactive presentation and analysis of the urban environment. TerraSim has been working on applications of its TerraTools® system to provide that support for several years. This article discusses some of the considerations involved in constructing a 3D world.

Remote Viewing

At its most basic level, 3D visualization allows us to view a site remotely. Used to its fullest potential, it can be employed as an urban planning tool, to view things as they might be, such as a new 40 story skyscraper where a 10 story building now stands, or to visualize relationships between urban structures in ways impractical in the real world, such as connections to underground utility and subway systems.

Urban planning is intimately concerned with the relationships among buildings, streets, and neighborhoods. A 3D visualization makes the relationships explicit, allowing careful consideration of the effects of a change in the urban fabric. Visualizing a proposed development within its urban context provides a precise idea of its impact on the neighborhood and interplay between adjacent structures. A 3D visualization greatly improves the public's understanding of a proposed development and facilitates communications between all parties involved, including:

Another important aspect of urban planning concerns emergency response, for fire or police calls or for disaster response/relief. A 3D simulation of the city can serve as a detailed planning map, a virtual training ground for scenario rehearsal, or as a situation monitoring display, should an emergency occur. Effective counter-terrorism operations require coordination between numerous government agencies and private organizations, which can be facilitated by an accurate shared model of the affected site.

Figure 1: 3D visualization of Philadelphia, near City Hall

Building the virtual world

Many different types of computer-generated scenes are described as “virtual worlds”, leading to confusion over data and software requirements and the quality of the visualization experience. Visualizations can be divided into four classes, roughly in order of complexity and correspondingly, visualization quality.

Constructing a realistic visualization of a complex urban area has, until now, been an expensive undertaking due to two main factors: the lack of suitable input data, and the construction cost due to limitations in the available tools. However, the increasing availability of more detailed data sets, and more importantly, major improvements in tools, processing algorithms, and graphics hardware, have made the construction of 3D urban visualizations a realistic goal.

The lack of sufficient input data is still a limitation on visualization construction. Both 3D geometry and appearance information are needed, but in some cases a GIS contains only 2D geometry and no appearance information. It follows that an effective toolset for 3D visualization must include a full suite of tools to model 3D objects and features from 2D data, as TerraTools does.

2D features such as roads may have no associated elevations, so they must be dropped to the DEM surface and integrated into the terrain. For 3D features represented in 2D, there are many different ways of turning the 2D data into 3D. A simple extrusion can be sufficient if attributes such as height are available. Geometry or appearance information can often be inferred from the attribution, if the construction tool has sufficient capabilities.

Buildings represented by 2D footprints/rooflines must be extruded to form a 3D volume, a simple process if the building heights are stored in the GIS data. Otherwise, heights may be inferred from other attributes. For instance, if building type or zoning information is available, residential buildings may be assumed to be two stories and commercial buildings three stories. Similar procedures may be used to assign building appearances.

For the visualization to be realistic at street level, it must contain the utility poles, street signs, traffic signals, and “street furniture” that populate city streets. Placing models at the GIS point locations given for these objects is straightforward, but determining the proper model orientation requires sophisticated reasoning about adjacent objects. For example, streetlights must be oriented so that the light is over the street and traffic signs must face the flow of traffic. In other cases, objects not normally included in a GIS, such as newspaper boxes and trash cans, may be placed at their typical locations to provide a more realistic street appearance. 3D geometry can also be generated based on standard geometric templates. For instance, curbs and sidewalks can be generated in TerraTools with specified widths, heights and materials.

Figure 2: “Street furniture” automatically placed from GIS source data

In addition to modeling considerations, an urban visualization tool must often handle a heterogeneous mix of data types. In some cases, standard cartographic or DEM data from the U.S. Geological Survey may be required to provide the context around a smaller area of interest for which GIS or CAD data is available. CAD data for 2D planimetrics or 3D models, most often in AutoCAD or MicroStation formats, may be available for specific sites or structures of interest, especially in planning applications. These different types of CAD data require different kinds of processing for real-time visualization.

Figure 3: From AutoCAD models to interactive site visualization

A limitation of many urban visualizations is that they do not allow the viewer to step off the street and into a building. The CAD data necessary to support the construction of the interior floors, stairways, doors, and windows is seldom available and, even when it is, is often not suitable for use in real-time visualization due to its high level of detail and lack of topology. TerraTools allows for automatic generation of plausible building interiors, given the building type, exterior building shape, and general parameters describing the desired interior.

Much of a city's complexity is hidden underground, with utilities, drainage, and transportation forming a complex hidden 3D web. Including this belowground infrastructure in the urban visualization allows it to be seen and understood without excavation, in relation to aboveground structures. The ability to inspect and manage underground assets with an interactive 3D visualization is a cost-effective alternative to excavation for water departments, electric companies, metropolitan transportation authorities, and telephone and cable providers.

Figure 4: Above and below ground: modeling subway platforms and tunnels

Traveling the virtual world

The viewer's sense of reality in a virtual world is determined both by the realism of the scene and their interactivity with it. The realism of the scene is (roughly) determined by the number of polygons included, with more polygons implying a more realistic rendering. However, realistic interactivity requires that the scene have as few polygons as possible so that graphics hardware can maintain realistic rates of motion through the scene.

Therefore, every construction step must be taken with its effect on viewing in mind---the number of polygons, the amount of texture memory required, and disk I/O, to name a few. As graphics hardware becomes more powerful the numbers change, but the problem remains the same. Increased graphics capability can be used either to render an un-optimized visualization, produced with less effort, or to render a more realistic or detailed visualization which has been optimized for display. Given this choice, we have found that users of 3D visualization technology always prefer to construct virtual environments that are as faithful to reality as possible, implying that highly optimized construction is a critical component of a 3D visualization toolset.

The biggest consideration is interactive vs. pre-planned viewing: whether the viewer can freely fly or drive through the scene, or is restricted to viewing non-interactive videos produced off-line. Off-line rendering can produce very realistic views, since the polygon count can be high, but these renderings are not interactive, eliminating the possibility of seeing the scene from another viewpoint, or traversing another path. In addition, off-line renderings may take hours or even days to complete.

Smarter database design circumvents the polygon count problem. More efficient representations, such as triangulated irregular networks (TIN), maintain the appearance of the terrain without adding unnecessary polygons. The fact that less detail is visible on an object, as one moves away from it is also advantageous. By including levels-of-detail (LODs) in the database and support for them in the associated viewer, less-detailed versions of models can be automatically substituted as the user moves farther away in the virtual world, providing the illusion of detail without sacrificing performance. TerraTools uses these methods to manage dense urban environments while maintaining real-time viewing rates.

Beyond visualization

While looking at and walking through a virtual world is useful for many applications, the underlying GIS data and associated databases often contain additional information which would be much more valuable if it were accessible through the 3D visualization. Imagine being able to click on a building in a visualization to obtain construction history, fire marshal's inspection reports, its tax appraisal records, or a directory of its occupants. TerraSim’s GISLink product delivers this capability.

GISLinkTM technology ties together the powerful information access capabilities of a GIS database and the intuitive interaction of a 3D visualization. In conjunction with the ITspatial VIO-GISTM viewer, GISLink enables users to seamlessly interact with both a standard GIS (ArcView) and a 3D visualization of the same scene. When features are selected in the 2D window, the 3D view flies to that point and the features are highlighted by a flashing box. Alternatively, objects selected in the 3D window are highlighted in the 2D view. Object selection can also be accomplished using standard GIS queries, since VIO-GIS is built on ArcView. This enables ArcView users to exploit 3D visualization tools immediately, without a steep learning curve or costly training in an unfamiliar toolset.

Figure 5: Linking GIS tools with real-time 3D visualization

For example, a user could drive down a virtual street and select a building in the visualization to query its tax assessment information. Conversely, they could write a GIS query to select all utility poles with a certain type of transformer, and then fly to the location of each one to examine surrounding buildings or terrain.

Visualizing the future

3D urban visualization is now at the same relative point in its development that GIS technology was several years ago. Sufficiently powerful PC graphics hardware is already widely and inexpensively available, due to graphics card developments fueled by the video game market; as capable software tools such as TerraTools become available and more powerful, the user community will grow rapidly. City planners can interactively judge the impact of proposed developments; commercial developers can communicate site plans dynamically to interested constituencies to facilitate project acceptance; utility companies can manage existing underground infrastructure; and multi-agency crisis response teams can rapidly form and execute plans in a high-fidelity 3D geospatial visualization. We at TerraSim look forward to the next few years.

Acknowledgements

The Philadelphia database shown was built by the Institute for Defense Analyses (IDA).

Author and Copyright Information

Copyright 2002 by author

William J. Starmer is a Geographic Technologies Specialist at TerraSim, where he is in charge of the Database Construction Group and also provides customer training and support. He has over eight years of experience in GIS and remote sensing and is currently completing his M.A. degree in Geography at the University of California Santa Barbara (UCSB).

Jefferey A. Shufelt is Director of Corporate Research and a co-founder of TerraSim. Dr. Shufelt supervises the research and development of advanced visualization tools at TerraSim, and serves as Principal Investigator on several SBIRs focused on automated urban modeling tools. Dr. Shufelt has over thirteen years of experience in cartographic feature detection and extraction, and received his Ph.D. in Computer Science from Carnegie Mellon University in 1996.

David M. McKeown is the President and a co-founder of TerraSim. He is also a Principal Research Computer Scientist at the Computer Science Department of Carnegie Mellon University, Pittsburgh PA. Mr. McKeown has over 25 years of experience in the areas of image understanding for the automated analysis of remotely sensed imagery, digital mapping and image/map database systems, and geospatial visualization. Mr. McKeown was awarded the 1995 Fairchild Award (Photogrammetry) from the American Society for Photogrammetry and Remote Sensing (ASPRS) and the 1996 Schermerhorn Award from the International Society of Photogrammetry and Remote Sensing (ISPRS).
The authors can be contacted at TerraSim, info@terrasim.com.