AUGMENTING 3D CITY MODELS WITH VISUALIZATION

OF REAL-TIME METEOROLOGICAL PHENOMENA

Frank Steinicke, J

org Mensmann, Klaus Hinrichs, Kai Rothaus

Institute of Computer Science, University of M

unster, Einsteinstr. 62, 48149 M

unster, Germany

Jan de Buhr, Antonio Kr

uger

Institute for Geoinformatics, University of M

unster, Robert-Koch-Str. 26, 48149 M

unster, Germany

eywords:

Interactive environments, augmented reality, weather visualization.

Abstract:

General interest in visualizations of digital 3D city models is growing rapidly, and several applications are

already available that display such models very realistically. Many authors have emphasized the importance

of the effects of realistic illumination for computer generated images, and this applies especially to the context

of 3D city visualization. However, current 3D city visualization applications rarely implement techniques for

achieving realistic illumination, in particular the effects caused by current weather-related phenomena. At

most, some geospatial visualization systems render artiﬁcial skies—sometimes with a georeferenced determi-

nation of the sun position—to give the user the impression of a real sky. However, such artiﬁcial renderings

are not sufﬁcient for real simulation purposes.

In this paper we present techniques to augment visualizations of digital 3D city models with real-time display

of georeferenced meteorological phenomena. For this purpose we retrieve weather information from different

sources, i. e., real-time images from cameras and radar data from web-based weather services, and we use

this information in the rendering process for realistic visualization of different weather-related issues, such as

clouds, rain, fog, etc. Our approach is not limited to a speciﬁc setup, and we have evaluated the results in a

user study presented in this paper.

1 INTRODUCTION

Availability of digital 3D city models is on the rise,

and more and more applications for visualizing such

models are forthcoming, ranging from Google Earth

and Microsoft Virtual Earth to academic or indus-

trial visualization systems (Beck, 2003; Dodge et al.,

1998; Steinicke et al., 2006). Although the amount

of data has increased dramatically, these visualiza-

tion and simulation applications can provide very de-

tailed views of a city. Often digital 3D city mod-

els are derived from cadastral maps available as two-

dimensional CAD data sets which include each build-

ing’s footprint as well as the number of ﬂoors and

the height between ﬂoors. In addition, information

about the infrastructure, street furniture and vegeta-

tion, and also aerial photographs or satellite pictures

may be available. Usually facade images as well

as some highly-detailed 3D models of architecturally

prominent buildings, so-called 3D landmarks, are in-

tegrated. All these features contribute to the realism

of a visualized city model.

Currently, most of the systems for visualization

of urban areas concentrate on displaying the earth’s

surface including the objects located on the surface,

i. e., the terrain with buildings, roads, trees, etc.; for

instance Google Earth displays highly-detailed and

textured terrain models onto which city models are

placed. Although many authors have emphasized the

importance of the effects of realistic illumination for

computer generated images, techniques for achiev-

ing realistic illumination are rarely applied in current

3D city visualization applications. Research is fo-

cussed on improving rendering of the ground surface

instead, e. g., using different caching strategies for

huge amounts of terrain data (Ropinski et al., 2007),

whereas rendering with realistic lighting and the im-

pact on the visualization of the city model is rarely

considered. For instance, when planning a new build-

ing it is important from where and with what inten-

359

Steinicke F., Mensmann J., Hinrichs K., Rothaus K., de Buhr J. and Krüger A. (2008).

AUGMENTING 3D CITY MODELS WITH VISUALIZATION OF REAL-TIME METEOROLOGICAL PHENOMENA.

In Proceedings of the Third International Conference on Computer Graphics Theory and Applications, pages 359-366

DOI: 10.5220/0001098003590366

 SciTePress

sity the sun hits the outer walls and the roof in or-

der to optimize energy consumption. Another exam-

ple is the problem to position a patio or a balcony in

such a way that it has as much sun exposure as pos-

sible. Some approaches have been presented which

consider the effects of light and shadow in city mod-

els, but they are focussed on light interaction between

virtual buildings only, without considering global ef-

fects caused by the virtual sky (D

ollner et al., 2006).

In this paper we present techniques to augment

visualizations of city models with real-time meteo-

rological phenomena. We include information from

real-time video captured with web-based or special-

ized weather cameras, and retrieve current weather in-

formation from web-services, for example, data about

precipitation, cloudiness or fog. Combining all this

information we render a virtual sky having a visual

appearance that comes close to that of the real sky,

and hence the perception of the current weather for a

virtual city model is improved.

The remainder of the paper is structured as fol-

lows. Section 2 examines work related to visualiza-

tion of meteorological phenomena. In Section 3 we

describe our hardware setup and the data acquisition

from different sources about the past, current and up-

coming weather. Furthermore, we explain how to in-

terpret and combine this data. This information is

used by the rendering techniques described in Sec-

tion 4 to generate a realistic sky. Section 5 presents a

user study in which we evaluate how close the visual

appearance of generated virtual skies comes to their

real counterparts. The paper concludes in Section 6.

2 RELATED WORK

Visualizing physical phenomena, e. g., atmospheric

scattering, is one of the main research topics in com-

puter graphics (Schafhitzel et al., 2007). Most of the

early work is based on ray tracing, which is an ap-

propriate method for creating photo-realistic repre-

sentations of the atmosphere. However, due to the

high computational costs, an interactive visualization

was not possible at that time and therefore other ap-

proaches had to be applied. In most systems for visu-

alizing 3D city models the sky is modeled with spher-

ical or cubical shapes surrounding the virtual camera.

Textures showing corresponding images of a sky are

added to the shapes and give the impression of a sky.

When using a sky box, the virtual sky is composed

of six images forming the sides of a cube, i. e., left,

right, top, bottom, front and back textures (see Fig-

ure 1(a)). With such a conﬁguration the appearance

of the virtual sky is limited to an inﬁnite distant sky,

(a) Skybox textures (b) Skydome texture

Figure 1: Texture-based approaches for generating static

virtual skies.

and moreover animations like moving clouds cannot

be integrated easily. Alternatively one can use spheri-

cal domes to which distorted textures are applied (see

Figure 1(b)). Multiple transparent sky domes can vi-

sualize different layers of clouds and stars.

In general, the textures applied to certain sky

geometries already include meteorological informa-

tion, such as color transitions, virtual clouds or a vir-

tual sun. Since static images are used, the visual ap-

pearance is restricted in such a way that changes of

the sky caused, for instance by winds etc., cannot be

incorporated. In order to allow smooth changes in the

visualization of the sky, geodesic domes composed

of uniform patches can be used. A geodesic dome

is an almost spherical structure based on a network

of patches that lie approximately on the surface of a

sphere. Each patch is deﬁned by a certain number of

vertices (at least three), whose color can be changed

in real-time.

Many authors focus on certain properties such as

sunlight or scattering (Riley et al., 2004). Clouds,

for instance, can be modeled as volumetric objects, as

ﬂat images, so-called 3D impostor projected on a sky

dome, or using hybrid approaches (Harris and Las-

tra, 2001; Wang, 2004). Further rendering techniques

for meteorological features have been proposed, such

as using particle systems (Reeves, 1983) for rain ren-

dering (Tatarchuk, 2006) or for visualizing falling

and accumulating snow (Sims, 1990; Fearing, 2000).

Most of these features can be realized using program-

mable shaders in order to allow simple and efﬁcient

rendering on the GPU (Roden and Parberry, 2005).

Some of these aspects have been included in sky ren-

dering libraries, such as SilverLining (Sundog Soft-

ware, 2007) or SkyWorks (Harris, 2004). However,

while these systems provide a good-looking impres-

sion of a virtual sky, they are more targeted on high

rendering performance than on realism. In particular

these applications do not take into account real-time

weather information in order to present an updated

sky.

The information required for rendering such a sky

can be retrieved from different sources. Reﬁned data

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

360

about general weather conditions is available via sev-

eral web-services. Satellite images comprise cloud

coverage and rainfall, but lack resolution for the rel-

atively small area covered by a city. Stationary cam-

eras that capture the sky provide a more local data

source. The obvious approach to augment the visual-

ization of a 3D city model is to use the resulting video

live streams as background information for the visu-

alization. However, a direct mapping of these images

is not sufﬁcient since they are limited to speciﬁc po-

sitions, and usually these cameras cannot record the

entire sky. The latter problem could be diminished

by using all-sky cameras (FMI, 2007) which are used

in astronomy and meteorology for studying aurorae,

meteors and cloud coverage. They are based on ﬁsh-

eye lenses or spherical mirrors and can observe the

whole sky at once due to their very wide ﬁeld of view

(FoV). However, the resulting distortion and their lim-

ited availability makes all-sky cameras unsuitable for

visualization purposes. Another problem with a sta-

tionary camera occurs when the user navigates the vir-

tual camera in such a way that the visible virtual sky

corresponds to a real sky which is out of range of the

stationary camera, and thereupon no data is available

for rendering the sky.

Our goal is to provide a rendered sky with a vi-

sual appearance that comes close to the real sky at the

present time. Furthermore, it should be possible to

visualize weather forecasts as well as weather history.

The virtual skies should be augmented by real-time

information from the acquired camera images as well

as information obtained from other discrete sources

such as weather radars.

3 DATA RETRIEVAL AND

ANALYSIS

In this section we describe the different sources from

where we retrieve the data, in particular current

weather information, in order to generate a virtual city

model and augment it with a realistically looking vir-

tual real-time sky.

3.1 Digital 3D City Model

In cooperation with the urban development, city plan-

ning and transport planning ofﬁce as well as the

land surveying and land registry ofﬁce of the city

of M

unster in Germany, we have developed a semi-

automatic process to generate a digital 3D city model

from the cadastral data provided for a city. The cadas-

tral data include for each building its footprint as

well as the number of ﬂoors and the height between

Figure 2: Virtual city model combined with a virtual sky

box showing computer generated static images.

ﬂoors, position and type of street furniture, informa-

tion about vegetation, etc. Moreover, the data pro-

vides an elevation model with corresponding aerial

photographs captured during overﬂights. For the dig-

ital 3D model a solid representing a building is gen-

erated by lifting the polygon describing the building’s

footprint by its total height which is calculated from

its number of ﬂoors and the height between ﬂoors.

Since a building’s roof type as well as information

about its facades are not included in the cadastral

data we apply heuristics in order to generate a pro-

totypical appearance of these features. In addition

we apply shader-based clipmapping-techniques in or-

der to display the enormous amount of areal images,

which include overall 120 GB of uncompressed data

(Ropinski et al., 2007). This technique enables us

to visualize shadows of clouds on the ground as de-

scribed in Section 4. Since automatically process-

ing virtual buildings as described above lacks most

of the details of their real counterparts particularly

for architecturally prominent buildings like churches.

For this reason we have modeled several of these 3D

landmarks with a very high level-of-detail manually

(Steinicke et al., 2006). The entire model consists

of more than 150,000 buildings, vegetation objects

and street furniture as well as more than 50 landmark

buildings. Besides the concepts described in this pa-

per, the model can also be enhanced with a sky based

on textured virtual skyboxes or virtual domes as de-

scribed in Section 2. But as mentioned above this

procedure is limited to static skies, whereas no cur-

rent meteorological information can be incorporated.

The image shown in Figure 2 illustrates the virtual

city model of the city of M

unster. The sky in this

image is displayed using a virtual sky box with the

textures from Figure 1(a).

AUGMENTING 3D CITY MODELS WITH VISUALIZATION OF REAL-TIME METEOROLOGICAL PHENOMENA

361

Figure 3: Image of the MOBOTIX HiRes IP camera.

3.2 Steerable Weather-Camera

We have installed a MOBOTIX network camera

(model M12M-Web-D43) on the roof of an eight-

storied building at the coordinates 51

◦



58.5



◦



11.5



E (see Figure 3) (Gliet, 2007). The cam-

era provides a resolution of 1280 × 960 pixels with

a wide angle lens (43 mm, F=2.0). It can be steered

360 degrees horizontally, whereas the vertical pitch is

about 80 degrees. Live stream from the camera and

further information is available via a web-based inter-

face

Figure 4 shows an image captured with the camera

during the summer in July about noon. Although this

camera provides good images for our approach, the

concepts are not restricted to this speciﬁc model. Any

camera or even static images may be used to interpret

the sky and get information as described below. Al-

ternatively we provide a GUI via which the user can

specify the parameters in an arbitrary way.

A trivial approach to generate a virtual sky from

the acquired camera images would be to extract only

the sky and to integrate this video stream into the vir-

tual city model. However, besides image quality prob-

lems due to restricted resolution as well as color falsi-

ﬁcation, the limited area which can be captured with

the camera prevents this approach since it results in

insufﬁcient images.

Hence, we choose a different approach. We apply

computer vision strategies in order to extract informa-

tion from images that show parts of the sky. From a

computer vision perspective the given task is to seg-

ment an outdoor image into three main components:

the sky, the clouds and the ground. In fact the latter

is not relevant itself, but is used for the detection of

the horizon and the skyline. We detect the horizon

(see Figure 4) and crop the image such that only the

sky is left. Thereupon we classify the sky pixels as

http://ifgicam2.uni-muenster.de

cloud or background. On this clustering result further

parameters for the proposed weather rendering tech-

nique could be deducted. We assume the horizon as

a vertical line and use an extension of the method of

Ettinger et al. (2003) for the detection of the horizon

line. The horizon is an important reference line. With

known internal camera parameters and known global

position of the weather camera, the captured pixels

are embedded in a real world reference system. Et-

tinger et al. classify sky and ground pixel by minimiz-

ing the inner variances of the two RGB-distributions.

We adopt their idea and formulate the optimization

criterion

J =

· min{S(Σ

),S(Σ

)} +

· max{S(Σ

), S(Σ

)}

with the score function S (Σ) := |Σ|+trace (Σ)

. Here

and Σ

are the class-dependent (sky and ground)

3 × 3 covariance matrices of the red, green and blue

channels. After detecting the horizon line (see Fig-

ure 4(a)), the skyline is extracted by a contour track-

ing method (based on a dynamic programming algo-

rithm) and an objective function similar to the energy

term J. We crop the sky part of the image for further

processing (see Figure 4(b)).

The next step is to classify the sky pixels as clouds

cloud

or sky background C

sky

. For this we utilize the

observation that under normal conditions the back-

ground is colored whereas the clouds are rarely col-

ored. The HSI-color model is adequately suited for

such a classifying task, since the S-value represents

the saturation of a pixel p. We use two thresholds t

low

and t

high

for classiﬁcation and the following rule:

p →

⎧

⎨

⎩

cloud

, if S(p) < t

low

sky

, if S(p) > t

high

both

, otherwise

(1)

The nonspeciﬁc class C

both

is introduced due to the

fact that clouds sometimes do not hold a sharp bound-

ary. At the boundary this kind of clouds are trans-

parent and the sky background is visible. Figure

4(c) shows the extracted sky background with origi-

nal color value (H), full saturation (S) and full inten-

sity (I) and the speciﬁc clouds in white. The inten-

sity value (I) of the clouds are visible in Figure 4(d).

Here the sky is represented in black. To compute the

percentage of sky cover, we calculate the observed

frequency of cloud pixels in relation to the complete

sky region. The colorization of background can be

estimated by averaging the sky background color de-

pending on the vertical FoV. Furthermore, if individ-

ual clouds can be segmented, i. e., they have a visible

boundary, we copy such clouds into the correspond-

ing textures which are applied during the rendering

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

362

(a) (b) (c) (d)

Figure 4: The images show the computer vision approach to retrieve information about the current weather: (a) original image

with detected horizon, (b) cropped sky region, (c) pure color of sky pixel with saturation and (d) intensity of cloud pixel with

saturation ≤ t

high

process for certain cloud structures (see Section 4).

The entire process can be performed interactively.

3.3 Weather Radar and Web-based

Weather Services

Recognizing certain weather-related features from a

photograph of the sky is limited. For example, solely

by using computer vision algorithms it is often dif-

ﬁcult to distinguish clouds from background such as

in Figure 6. Rain or snow can be distinguished only

when precipitation is extremely high. Currently, we

are working on a more ﬂexible classiﬁcation method

for cloud and background pixels, which can operate

in more general conditions (dusk, dawn, back light,

halos and so on).

However, due to these limitations we extract fur-

ther weather-related information from weather radar

and web-based weather services. Our cooperation

partner, the working group “climatology” of the Insti-

tute for Landscape Ecology, provides these services to

retrieve data (Institute for Landscape Ecology, M

uns-

ter, 2007) as well as to control and steer sensors, e. g.,

the weather camera. These services provide informa-

tion about minimum and maximum temperature, di-

rection and strength of wind, time of sunrise and sun-

set and percentage of possible sunshine, precipitation

probability and amount of precipitation, atmospheric

humidity, etc. This sensor data is encapsulated by

standardized web services, where three services spec-

iﬁed by the Open Geospatial Consortium (OGC) are

used (Open Geospatial Consortium, 2007):

1. sensor observation services (SOS) which retrieve

meteorological data (e. g. wind direction, wind

speed, temperature),

2. web coverage services (WCS) that retrieve georef-

erenced weather radar data, and

3. sensor planning service (SPS) which enable to

steer the camera.

Data that is obtained from the weather services re-

spectively the weather radar is compared to the inter-

pretation of the camera images. When both differ, for

example, the camera interpretation yields 80% cloud

cover, whereas the weather service prognoses 70%,

we average the results and tune the parameters such

that the visualization of the virtual sky contains 75%

clouds.

4 RENDERING TECHNIQUES

FOR METEOROLOGICAL

PHENOMENA

In this section we explain different rendering tech-

niques that we use for the described meteorological

phenomena.

4.1 Atmospheric Rendering

In our approach atmospheric rendering is based on

a layered model composed of three geodesic domes.

Each geodesic dome is composed of uniformly sepa-

rated patches each having nearly the same size. This

ensures that the sky shading is smooth across the en-

tire dome surface. As depicted in Figure 5 we position

the virtual sun represented by a virtual sphere on the

surface of the outer dome with respect to the georef-

erenced position of the city model as well as the local

time. The size as well as the color is determined by

values derived from the weather web-service. In addi-

tion a distance light is set with respect to the sun’s po-

sition and directed to the center of the 3D city model.

The visualization of the atmosphere is simulated

by multiple intervals with linear color transitions,

deﬁning the color from the horizon to the zenith. The

color values are deﬁned with respect to the color val-

ues determined by the colors from pure color of the

sky images as depicted in Figure 4. Since the cam-

era’s orientation as well as the FoV that records the

image or video stream of the sky is known, corre-

sponding colors are assigned to the vertices of the

geodesic dome (see Figure 5). The vertices which

cannot be assigned to pixels in the camera images are

AUGMENTING 3D CITY MODELS WITH VISUALIZATION OF REAL-TIME METEOROLOGICAL PHENOMENA

363

high clouds dome

middle clouds dome

low clouds dome

example patch

camera position

colored vertex

sun sphere

Figure 5: Arrangement of virtual skydomes supporting

three different layers of clouds.

colored with respect to their distances to already col-

ored vertices while the geodesic dome’s symmetry is

exploited, i. e., vertices which have the same height

are colored in the same way. In addition when ver-

tices are close to the virtual sun even the color of the

virtual sun is merged into the color of the vertex (see

Figure 5). Alternatively, the colors at the horizon and

in the zenith can be deﬁned by the user via a graphi-

cal user interface (GUI). Since the color assignment is

performed in programmable vertex shaders rendering

can be performed in real-time.

4.2 Rendering of Clouds

As described above the three geodesic domes are

composed of almost uniform patches. Consequently,

there is no singular point in which the patches con-

verge, and thus distortion of textures applied to the

surfaces is prevented. Each geodesic dome is associ-

ated to one family of clouds, i. e., high, middle and

low clouds, which are represented as textures on the

corresponding surface. According to (Howard, 1802)

we classify clouds by the altitude of the cloud base.

High clouds form above 5,000 meters in the cold re-

gion of the troposphere; they are denoted as cirrus

or by the preﬁx cirro-. The clouds tend to be wispy

and are often transparent, e. g., cirrus, cirrostratus,

cirrocumulus, etc. Middle clouds develop between

2,000 and 5,000 meters and are denoted by the preﬁx

alto-, for example, altostratus and altocumulus. Low

clouds are found up to 2,000 meters and include the

stratus (dense and grey). When stratus clouds con-

tact the ground, they are called fog (see Section 4.3).

Low clouds in this family include, for example, stra-

tus, nimbostratus and stratocumulus.

In order to be able to render clouds at differ-

ent altitudes realistically we render each family of

clouds on the corresponding geodesic’s surface (see

Figure 5). Since the approach is focussed on pedes-

trian navigation where clouds are reasonable far away

from the user’s perspective this approach yields to

sufﬁcient results (Wang, 2004). For each cloud fam-

ily we have generated texture atlases that are com-

posed of different cloud-based textures representing

different cloud types. The textures are encoded in the

RGBA format, i. e., each texel contains an RGB color

entry as well as a transparency value α. Each type of

cloud can automatically be switched on or off and the

amount of each cloud can be controlled as described

in Section 3.2. Alternatively clouds can be controlled

via the GUI by means of controlling the color chan-

nels. For instance, black is associated to no clouds of

the corresponding type, while white is associated to

all clouds represented in the texture. The transparency

of clouds is mapped directly from the α value in the

texture. Furthermore, the color of the clouds can be

changed by means of associating a grayscale value to

each α value. Since we use shaders for each cloud

family, clouds from the same as well as from differ-

ent families can be combined arbitrarily in real-time.

Wind information that we retrieve from the

weather web-service is used to move the clouds in

the corresponding direction by means of rotating each

geodesic dome. The rotation axes are determined by

the cross product of the wind direction and the nor-

mal at the geo-referenced point to the ground model.

Due to the different radii of the three geodesic domes

we ensure that motion parallax caused by clouds is

analog to the real world, i. e., lower clouds appear

to move faster than middle clouds, the latter appear

to move move faster than high clouds. The rotation

speed and axis of each dome is determined by the

wind speed. Furthermore, we use the cloud textures

to apply corresponding shadows on the ground with

respect to the sun position.

4.3 Rendering of Rain and Snow

In our system techniques to render rain and snow are

similar to the rendering of clouds. Different textures

represent different kinds of snow or rain, which are

classiﬁed in corresponding categories again. For ex-

ample, according to the amount of precipitation, rain

can range from very light rain, i. e., the precipitation

rate is < 0.25 mm/hour, to moderate rain—when the

precipitation rate is between 1.0 and 4.0 mm/hour,

and to extreme rain, where the precipitation rate is

> 50.0 mm/hour. For each type of rain respectively

snow textures are combined in texture atlases again.

These textures are blended to the inner dome as well

as to the near clipping plane. By means of modifying

the texel position vertically the user gets the notion

of falling rain respectively snow. However, currently

rain and snow have no impact on the visualization of

the ground, e. g., accumulation of fallen snow. Due to

limitations in the computer vision process, we extract

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

364

information about rain or snow via the weather radar

only.

4.4 Rendering of Fog

The weather web-service gives information about the

visibility range determined by the level of fog. Fog

rendering is implemented in a fragment shader ap-

plied to each fragment, where transparent grayscale

values are blended with respect to its depth value,

i. e., objects which are far away from the camera are

blended with less transparency than objects close to

the camera.

5 SUBJECTIVE EVALUATION

In this section we present an evaluation in which users

had to evaluate how close the visualization of virtual

skies come to their real counterparts.

5.1 Tasks

We have prepared photographs which show a city

from different locations with different weather con-

ditions (see left images in Figure 6). After extract-

ing the corresponding information to deﬁne parame-

ters for the rendering process for the virtual sky as

described in Section 3, we have rendered an image

of the virtual 3D city from the same georeferenced

position enhanced with a virtual sky based on our ap-

proaches. We have rendered four more images from

the same position, but have applied a quasi-randomly

(QR) generated sky, i. e., we restricted the visualiza-

tion of that sky such that almost the same weather is

shown, for instance, if the photograph does not show

any clouds, we omitted clouds in the random process.

After comparing each rendered image side-by-

side with the original photo 6 (5 male, 1 female) par-

ticipant students had to state how close the visualiza-

tion of the virtual skies mirror the real ones in terms

of weather-related phenomena. The evaluation had to

be performed on a ten point Likert-scale, where 1 cor-

responds to ”no similarity at all” and 10 corresponds

to ”equality”, i. e., the images show the same weather.

Thereby similarity of the virtual and real weather in

terms of a subjective evaluation is possible, whereas

a comparison based on HDR images would result in

objective evaluation.

5.2 Results

The results of the user evaluation are illustrated in Ta-

ble 1. The table shows the average score of how users

reported the similarity of the sky visualized using our

concepts in contrast to using the QR generated images

with the corresponding photographs.

Table 1: Evaluation of rendered images with respect to pho-

tographs.

Image: Score (φ):

Fig. 6 (a) (right) 7.2

QR image of Fig. 6 (a) (left) 4.6

Fig. 6 (b) (right) 8.2

QR image of Fig. 6 (a) (left) 4.8

Fig. 6 (c) (right) 8.1

QR image of Fig. 6 (a) (left) 5.2

Fig. 6 (d) (right) 8.4

QR image of Fig. 6 (a) (left) 4.45

The results show that users perceive the visualiza-

tion of a virtual sky generated with our approach close

to the real sky, for example, depicted in a photograph.

Overall, users rated our concepts with 7.9 on average,

while a QR image of the same weather has been rated

with 4.76 on average.

6 CONCLUSIONS

We have presented techniques to enhance visualiza-

tions of digital 3D city models with real-time display

of georeferenced meteorological phenomena, such as

the visualization of atmosphere, clouds, sun, rain,

snow and fog. In contrast to other sky rendering li-

braries we have integrated real-time information re-

trieved from different sources, i. e., images from cam-

eras and data from web-based weather services, into

the rendering process in order to achieve a realistic vi-

sualization. The evaluation has proven that skies ren-

dered with our approach come close to the real skies

shown in photographs. Our approach is not limited to

a speciﬁc setup, i. e., the weather camera, but is ap-

plicable with any web-based camera as long as parts

of the sky are visible.

In the future we will develop an interface such that

that multiple images from the current sky can be com-

bined to further enhance the visualization of the sky.

Moreover, we will extend the computer vision process

in order to retrieve more information from each im-

age.

REFERENCES

Beck, M. (2003). Real-time visualization of big 3D city

models. International Archives of the Photogramme-

try, Remote Sensing and Spatial Information Sciences,

XXXIV(5/W10).

AUGMENTING 3D CITY MODELS WITH VISUALIZATION OF REAL-TIME METEOROLOGICAL PHENOMENA

365

(a) (b)

Figure 6: Visual comparison between the real respectively virtual sky in photographs respectively rendered images from (a),

(b) pedestrian perspective and from (c), (d) bird’s-eye view.

Dodge, M., Doyle, S., Smith, A., and Fleetwood, S. (1998).

Towards the virtual city: VR & internet GIS for urban

planning. In Workshop on Virtual Reality and Geo-

graphical Information Systems.

ollner, J., Buchholz, H., and Lorenz, H. (2006). Ambient

Occlusion - Ein Schritt zur Realistischen Beleuchtung

von 3D-Stadtmodellen. In GIS - Zeitschrift f

ur Geoin-

formatik, pages 7–13.

Fearing, P. (2000). Computer modelling of fallen snow.

In SIGGRAPH ’00: Proceedings of the 27th an-

nual conference on Computer graphics and interac-

tive techniques, pages 37–46. ACM Press/Addison-

Wesley Publishing Co.

FMI (2007). Finnish Metereological Institute all-sky cam-

eras. http://www.ava.fmi.ﬁ/MIRACLE/ASC.

Gliet, J. (2007). Augmented reality weather cam. Master’s

thesis, Institute of Geoinformatics, WWU M

unster.

Harris, M. (2004). SkyWorks cloud rendering engine.

htp://www.markmark.net/SkyWorks.

Harris, M. and Lastra, A. (2001). Real-time cloud render-

ing. Computer Graphics Forum, 20(3).

Howard, L. (1802). The modiﬁcation of clouds. Presenta-

tion to the Askesian Society.

Institute for Landscape Ecology, M

unster (2007). Cur-

rent weather in M

unster. http://kli.uni-muenster.de/en/

weather.

Open Geospatial Consortium (2007). OpenGIS speciﬁca-

tions. http://www.opengeospatial.org/standards.

Reeves, W. T. (1983). Particle systems – a technique for

modeling a class of fuzzy objects. In SIGGRAPH ’83:

Proceedings of the 10th annual conference on Com-

puter graphics and interactive techniques, pages 359–

375. ACM Press.

Riley, K., Ebert, D., Kraus, M., Tessendorf, J., and Hansen,

C. (2004). Efﬁcient rendering of atmospheric phe-

nomena. In Proceedings of the 15th Eurographics

Workshop on Rendering Techniques, pages 374–386.

Roden, T. and Parberry, I. (2005). Clouds and stars: Ef-

ﬁcient real-time procedural sky rendering using 3D

hardware. In ACE ’05: Proceedings of the 2005

ACM SIGCHI International Conference on Advances

in computer entertainment technology, pages 434–

437. ACM Press.

Ropinski, T., Steinicke, F., Meyer-Spradow, J., and Hin-

richs, K. (2007). Automatic integration of fo-

liage into 3D city models. In International Confer-

ence on Computer Graphics Theory and Applications

(GRAPP2007), pages 299–304.

Schafhitzel, T., Falk, M., and Ertl, T. (2007). Real-time

rendering of planets with atmospheres. In Journal of

WSCG.

Sims, K. (1990). Particle animation and rendering using

data parallel computation. In SIGGRAPH ’90: Pro-

ceedings of the 17th annual conference on Computer

graphics and interactive techniques, pages 405–413.

ACM Press.

Steinicke, F., Ropinski, T., Hinrichs, K., and Mensmann, J.

(2006). Urban city planning in semi-immersive virtual

reality. In Proceedings of the International Confer-

ence on Computer Graphics Theory and Applications

(GRAPP2006), pages 192–199. INSTICC Press.

Sundog Software (2007). SilverLining: 3D cloud and sky

visual simulation. http://www.sundog-soft.com.

Tatarchuk, N. (2006). Artist-directable real-time rain ren-

dering in city environments. In SIGGRAPH ’06: ACM

SIGGRAPH 2006 Courses, pages 23–64. ACM Press.

Wang, N. (2004). Realistic and fast cloud rendering. Jour-

nal of Graphic Tools, 9(3):21–40.

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

366