© since 1995, Eckart Lange, IRL-Institut, ETHZürich


Reality andComputerized Visual Simulation:
An Empirical Study about the Degree of Realism of VirtualLandscapes



In planning rather abstract, two-dimensional representations arepredominantly used, although the real world is three-dimensional.Besides traditional analog visualization techniques recently digital3D-visualizations have also become more common. The validation ofsimulations of virtual landscapes regarding their degree of realismhas so far been neglected in research.
The research project concentrates on the question whether, how, andto which degree the real visually perceived landscape, representedthrough photographs, can be validly represented by digital virtuallandscapes. The study area comprises the communities of Schwyz andIngenbohl-Brunnen in Central Switzerland.

Visualizing the third dimension
Instead of manually modelling the virtual environment, which is thetraditional CAD-approach, a GIS-based approach is pursued. This isthe prerequisite for the efficient visualization of large data sets.The simulated abstraction of the real landscape is mostly relying ondata which is either already available or on future publiclyaccessable 2D-data which can be made available through largelyautomatic procedures. For the digital visual simulation a virtualmodel of the study area was assembled, consisting of various elementssuch as the DHM25 digital terrain model, LANDSAT TM and Swissphotohigh-resolution orthophoto imagery. The elements forests, singletrees and buildings are mostly derived from the topographic map 1 :25'000.

The experiment: Real and Virtual Landscape
The validity of the created virtual landscape is tested in anempirical study. Using scaling techniques test persons are asked toorder a set of 90 images, which are varying in detail, according tothe degree of realism on a scale ranging from 1 - 5 (very low - veryhigh). Three images are photographs of real scenes (abackground-scene, a middleground-scene and a foreground-scene).Another 86 images are corresponding computer-generated images withdifferent representation levels. One image is a composite of photoforeground and virtual background. The test set is evaluated by 75test persons, grouped in non-local experts, non-local lay persons,local experts, and local lay persons.

6 of 90 test images:

Brunnen/Schwyz Region: View from the Mythen (1899m)

R-Degree 4.747 ......................... R-Degree 4.307

View of Brunnen with Lake Lucerne

R-Degree 4.96 ........................... R-Degree3.013

View towards Schwyz and the Mythen (1899m)

R-Degree 4.827 ......................... R-Degree 3.48

Results of the experiment

The results gained through Analysis of Variance show, that thevariance of the dependent variable 'degree of realism' is explainedto a very high degree by the independent variables of the model, i.e.the elements of the virtual landscape. The explained variance rangesfrom 87% for the foreground-scene to 82% for the middleground-sceneand 97% for the background-scene. By far the most important variablepositively contributing to the degree of realism is the terrain withthe draped aerial orthophoto. Second most important is the variablebuildings. The evaluation is significantly influenced in a positiveway if texture-mapped built-form is displayed. For this step manualmodelling is needed.
Overall the background-scene is rated with the highest degree ofrealism with the real scene receiving a value of 4.747 and thehighest rated computer-generated image getting 4.307. The highestrating for the middleground-scene is 3.013. The foreground-scene getsa value of 2.578 without texture-mapped buildings and 3.48 withtexture-mapped buildings. The composite image is rated 4.4.

The highest-rated virtual landscapes reach a degree of realism whichmakes them very hard to be identifed as computer-generated imagery.In the experiment, approximately 75% of the test persons assign thehighest possible value (åvery high degree of realism’) toone or more scenes of simulated landscapes. In most cases it is abackground-scene, which is placed in this category.
The middleground- and the foreground-scene are generally rated lessrealistic than the background-scene. More detailed data would benecessary to achieve a higher degree of realism. With the exceptionof digital terrain and remote sensing data, most of the datacurrently available for 3D-visualization are only covering parts ofSwitzerland. In the future more 2D-data as well as 3D-data will bereadily available. This will be the basis for a shift from thetraditional 2D-planning towards 3D-planning which could be offeringimproved support for better decisions about the spatial organizationof the landscape. However, visualization is so far mainly understoodas a presentation medium for the result of the planning process. Anessential requirement for visualization to provide a substantialcontribution in planning is the comprehension of visualization as anintegrated and at the same time integrating part in the planningprocess, facilitating improved communication among experts and laypersons, i.e. among planners and the persons affected byplanning.


Back to Eckart Lange