Você está na página 1de 4

AUGMENTED REALITY: FUSING THE REAL AND SYNTHETIC WORLDS

Mauro Dalla Mura, Michele Zanin, Claudio Andreatta, Paul Chippendale

Technologies of Vision, Fondazione Bruno Kessler, Trento, Italy


{dallamura, mizanin, andreatta, chippendale}@fbk.eu

ABSTRACT System (GIS) domains [1, Ch. 33]. In [3], a GIS-based photo-
Augmented Reality (AR) offers a means to inject virtual in- realistic visualisation method for viewing natural landscapes
formation into real scenes. In the past few years, AR has through video-based off-line AR was proposed. ARVino, a
been receiving greater attention thanks to considerable ad- demonstrator for the augmented visualisation of geolocated
vancements in the hardware of consumer-level portable de- wine-growing data was presented in [4]. In [5] it was shown
vices. In this paper, we illustrate how mobile AR can be ex- how geo-data could be exploited for labeling mountains that
ploited to intuitively visualize, and moreover generate new, are really visible inside photos, adding their heights above
geo-data. We explore these two concepts through i) the vi- sea level and their distances from the camera. Nevertheless,
sualization and interaction modality of geo-data as AR layers a stable and uncluttered integration of AR technologies for
and ii) the exploitation of mobile devices as opportunistic sen- geoscience applications is still to be achieved.
sors for generating information relating to a users immediate In this paper, we present two reciprocal aspects relating to
surroundings. geoscience applications, in which AR technology can be ex-
ploited to link the real world to geo-data archives. In the rst
Index Terms Augmented reality, image registration, case, AR can be seen as a tool to enable a user to experience
visualization, opportunistic sensors, data collection. an immersive new modality for accessing and interacting with
geo-data through the visual overlaying of synthetic informa-
1. INTRODUCTION tion onto the real scene. In the second case, since convincing
AR applications require a very precise registration of the de-
Augmented Reality is a technology which enables the blend- vice with the viewed scene, information can be gleaned from
ing of virtual and real scenes in both time and space [1, Ch. 1], reality through the devices sensors and can thus be exploited
opening up the possibility to inject information into a scene for augmenting geo-databases.
when viewed through a portable device. AR can be useful in
many applicative domains [1], such as in touristic guidance  
(e.g., providing detailed descriptions of monuments), enter- 

tainment (e.g., visualizing information about the players in a
 

 
football match, or, overlaying interactive gaming characters
into a real scene), urban planning (e.g., rendering virtual con-
struction models into a site), medical applications (e.g., pro- 

  

 

   
 
viding assistance during surgical operations), etc..
The latest developments in AR technology (i.e., in terms
  
of software and hardware) are enabling applications to run
entirely on consumer level mobile devices (e.g., smartphones Fig. 1. AR concept applied to geoscience and remote sensing.
and tablets) in real-time [1, 2]. Obviously, this increases the
potential impact that such technologies can have in our real Figure 1 illustrates the way in which AR can be exploited
life due to their profusion. in the elds of geoscience and remote sensing, highlighting
Thanks to these recent advances and the availability of the proposed symbiotic relationship.
massive geographic referenced archives of data (e.g., remote
sensing, geo-referenced social photo websites and geoscience
data) an effective and meaningful integration of AR and geo- 2. REGISTERING THE REAL TO
data is becoming feasible. Recently, we have started to wit- THE SYNTHETIC WORLD
ness a convergence of AR and the Geographic Information
This research has been funded by the European 7th Framework Program, In order to provide an immersive and convincing experience,
under grant VENTURI (FP7-288238). AR relies on a precise, consistent, and solid registration

978-1-4673-1159-5/12/$31.00 2012 IEEE 170 IGARSS 2012


between the real scene and a synthetic representation of the process is presented. In this scheme, alignment is obtained
world [1]. This is essential for a precise and non-jittery in- by nding matches between fragments of an edge-map (gen-
sertion of markers or points of interest into the environment erated using image processing), to similar proles in a syn-
(e.g., accurate placement of toponyms or overlays). thetically rendered panorama created by ray-tracing [5]. In
The registration of real scenes to synthetic models can this example, the correlation of features is achieved through
be performed in different ways. Latest generation smart de- spherical cross-correlation [6].
vices are equipped with on-board hardware, such as Global Upon registration, rendered virtual content can be over-
Positioning System (GPS), magnetometer, gyroscope, and laid into a real scene, being subsequently adapted in response
accelerometer, that can provide a rough estimate of pose to camera pose changes (e.g., a users movements). These
and position. When geo-positional aware landmarks are in pose changes can be detected and tracked by exploiting on-
view (e.g., mountain peaks, distinctive architecture, anthropic board hardware sensors (e.g., accelerometers and gyros)
structures, etc.,) photo-to-world registration can be driven by and/or through the temporal monitoring of visual feature
vision-based features. displacements.
Vision-based registration is often obtained through the Using such a terrain-aware approach avoids the overlay-
cross-correlation of geo-referenceable synthetic features to ing of occluded content or the representation of details that
visually detectable naturally occurring ones [6]. In this ap- are not visible due to their distance from the viewing point.
proach, 3D-models are built around a users location (increas- Thus, in this way, a convincing blend between the real scene
ingly appearing inside the EXIF metadata of photos) [5, 6] and superimposed virtual information can be provided as nat-
from geo-data such as Digital Elevation Models (DEMs), urally as possible.
geo-referenced maps, and vectorial GIS layers.
As a consequence of registration, a direct mapping of 3. AUGMENTING THE USERS VIEW
land-based photo pixels to geographic coordinates can be
obtained. Once a real scene is geo-registered, it is possible to offer the
user intuitive access to many forms of geo-data.
Public and private geo-data (e.g., geographical toponyms,
GIS layers, 3D buildings models) represent potentially rich
sources of augmentation material that can easily be sup-
plied through web services. Since the superimposition of

   augmented overlays is subject to their actual visibility from
the users vantage point, this becomes a natural modality to
 
   query, lter and show only the information pertaining to the
   
users surroundings. Touchscreen and gesture-based interac-
tions, now commonly available in consumer-level devices,
can provide an intuitive interface to browse and manipulate
augmented content.
This means of information delivery can directly benet
the general public since it provides intuitive visualizations and
      enables simple geo-data operations; moreover, it can also pro-
vide powerful tools for geo-scientists.
It is well known that eld work campaigns are of fun-
  damental importance in remote sensing, however, they are
usually very time-consuming and often tedious [7]. In this
respect, AR scene augmentation provides a valuable tool to
aid operations such as the in-situ validation of remote sens-
ing products (e.g., land cover thematic maps). In the same
way, it can also provide real-time guidance for eld surveys
gathering reference data (e.g., ground truth collection). Fur-
 
 
thermore, a more accurate registration of the device perform-
ing the augmented overlays (e.g., using hardware and vision-
based approaches) can help to reduce location errors residing
Fig. 2. Flowchart of a vision-based photo-to-world registra- in geo-data, which is one of the issues that affects eld mea-
tion [5]. surements [7].
The interpretation of remote sensing data can also be aided by
In Figure 2 a typical pipeline of a vision-based registration scene augmentation. For example, by re-projecting and over-

171
laying geo-referenced satellite or aerial images on the real
scene can help users to associate features in an image to cor-
responding objects in the real world. This can be particularly
useful when dealing with active sensor data (e.g., SAR and
LIDAR), or optical data acquired at spectral wavelengths be-
yond the visible spectrum (e.g., hyperspectral images). This
in-situ visualization of geo-data, whether static or time-series,
enables a user to easily detect and track abrupt changes in the
landscape and facilitates the intuitive observation of land-use
evolution (e.g., culture transitions in agricultural elds). In
this scenario, archives of geo-tagged photos residing in large
web databases (e.g., Flickr and Panoramio), can also be ex-
ploited (once geo-referenced) to extend the spatio-temporal
coverage of a landscape.
In Figure 3 we show how an augmented layer can provide
the user with additional information about their surroundings,
illustrating the names of visible peaks, the path of a highway
and the course of a river. As this gure exemplies, it is a
straightforward procedure to intuitively verify the semantic
correctness and spatial precision of geo-information pertain-
ing to a region.

Fig. 3. View from Fai della Paganella (Trento, Italy) aug-


4. AUGMENTING THE GEO-DATA ARCHIVES mented with the toponyms of visible mountains peak, a high-
way (black line) and the Adige river (blue line).
Thanks to accurate real-worldsynthetic registration, the All of the geo-data used for the registration and augmentation
spatial coverage of each pixel from an aligned photo is is publicly available on the internet: toponyms were obtained
known. Thus, pixels from a ground-level image can eas- from GeoNames (http://www.geonames.org/), the
ily be re-projected into a nadiral view (i.e., a viewshed), DEM used for the registration is the SRTM 90m Digital Ele-
for generating new GIS layers or for comparisons to other vation Data distributed by CGIAR Consortium for Spatial In-
satellite images. formation (http://srtm.csi.cgiar.org/), and the
In this way, we can consider AR-enabled portable devices track of the highway and the river are taken from Open-
as ground-based opportunistic mobile sensors which are ex- StreetMap (http://www.openstreetmap.org/).
ploitable to acquire high spatial resolution information from Source http://tev.fbk.eu/marmota.
the scene. This can provide geoscientists with the ability to
create new, or to update existing geo-data archives. From
this perspective, through the collection and accurate interpre-
tation of geo-specic information registered by many users,
it is possible to augment, model and document an evolving
landscape. For example, the use of up-to-date and ground-
level visual evidence could be exploited in tasks such as the ral aerial view. Once re-projected, the new viewpoint can be
detection of changes in land usage with respect to previously rapidly compared to previously acquired remote sensed im-
available data, the monitoring of agricultural, forest and urban ages, providing hints on the evolution of a landscape. Need-
developments, the estimation of snow coverage, the extent of less to say, there are precision issues due to the spatial distor-
ooding and the generation of damage assessment maps. tion introduced by the uneven sampling of the landscape onto
Furthermore, since the acquisition is performed at the the ground (i.e., the spatial resolution varies according to the
ground level, the users immediate surroundings, as sensed distance from the camera, as can be noticed in Figure 4), or by
by a devices camera, can provide information at a very high the limited spatial coverage, or the presence of foreground ob-
level of precision, although limited geographically. This in- jects masking the landscape. Nevertheless, such social data
formation can be useful for describing the land cover type. sources might represent a valuable new resource of observa-
If multiple geo-registered views of the same area are avail- tional data since it comes at no cost. Furthermore, ground-
able, it is also possible to generate new 3D terrain datasets or based views of the environment can capture vertical surfaces,
building models using photogrammetry techniques. produce usually cloud-free images, and could represents the
Figure 4 illustrates the landscape in Figure 3 re-projected only near-live acquisition available for a disaster area (e.g., in
onto a top-down view as if it had been acquired from a nadi- a scenario of damage assessment).

172
The potential of hybrid systems that can harmonize infor-
mation from aerial and ground-based social sensors is only
just beginning to be explored, but this eld of research is
promising to accelerate the access and creation of shared and
comprehensive data records. In this scenario, Augmented
Reality has the potential of becoming a novel and inexpen-
sive tool for accessing and interacting with geo-data through
consumer-level portable devices. Furthermore, thanks to AR
technology that is able to precisely align a real scene with
geo-referenced 3D models, it is possible to think of mobile
imaging devices as remote sensors acquiring ground-level in-
formation. From this perspective, users devices can be ex-
ploited for the updating of geo-data archives in a collaborative
and participative way.

6. REFERENCES

[1] B. Furht, Handbook of Augmented Reality, Springer,


2011.
[2] M. Billinghurst and A. Henrysson, Mobile architectural
augmented reality, in Mixed Reality In Architecture, De-
sign And Construction, Xiangyu Wang and Marc Aurel
Fig. 4. Viewshed of photograph from Figure 3, re-projected Schnabel, Eds., pp. 93104. Springer Netherlands, 2009.
onto the ground surface. An equally spaced grid (in original
photo pixels) is overlaid onto the viewshed to illustrate un- [3] P. Ghadirian and I.D. Bishop, Integration of augmented
even spatial resolution due to re-projection. reality and GIS: a new approach to realistic landscape vi-
sualisation, Landscape and Urban Planning, vol. 86, no.
3-4, pp. 226232, 2008.
5. DISCUSSION AND CONCLUSION
[4] G.R. King, W. Piekarski, and B.H. Thomas, ARVino-
When augmenting real scenes and populating geo-data archives, outdoor augmented reality visualisation of viticulture
some aspects are essential in order to provide the user with a GIS data, in Mixed and Augmented Reality, 2005. Pro-
realistic experience and to generate accurate information. ceedings. Fourth IEEE and ACM International Sympo-
The re-projection of geo-data onto the users viewpoint sium on, 2005, pp. 5255.
for scene augmentation is typically a demanding op-
[5] P. Chippendale, M. Zanin, and C. Andreatta, Spatial and
eration and must be performed in a fast and accurate
temporal attractiveness analysis through geo-referenced
manner. Efcient rendering techniques for the fusion of
photo alignment, in IEEE International Geoscience and
real and synthetic information must be considered. The
Remote Sensing Symposium, 2008. IGARSS 2008, July
use of Graphics Processing Units (GPUs), which have
2008, vol. 2, pp. II1116 II1119.
already proved to be effective for performing fast ren-
dering and the processing of large amounts of remote [6] L. Baboud, M. Cadik, E. Eisemann, and H. P. Seidel,
sensing data [8], are promising to provide a solution to Automatic photo-to-terrain alignment for the annotation
this problem as GPUs are now available in many smart of mountain pictures, in IEEE Conference on Com-
devices. puter Vision and Pattern Recognition (CVPR), 2011, June
2011, pp. 4148.
Precise localization and device pose are essential com-
ponent to deliver accurate AR overlays. Joint vision- [7] R.M. McCoy, Field Methods In Remote Sensing, Guil-
and sensor-assisted approaches must be considered to ford Press, 2005.
increase the robustness of the registration chain.
[8] M. Lambers and A. Kolb, Gpu-based framework for dis-
High-speed wireless data-transfer speeds between tributed interactive 3d visualization of multimodal remote
portable devices and web servers are essential since sensing data, in Proc. Int. IEEE Geoscience and Remote
large amounts of geo-data need to be transferred back Sensing Symposium (IGARSS), 2009, vol. 4, pp. IV57
and forth. Network transfers can be optimized by con- IV60.
sidering compression algorithms and ofine storage.

173

Você também pode gostar