Communication Aspects: How Landmarks Enrich the Communication Between Human and Machine - Landmarks: GIScience for Intelligent Services - page 183

accompanied by a single photograph that highlights the most important visual
information for finding the way, for example, the entrance to the building. Lee
et al. [ 48 ] integrated photographs of landmark buildings in a map view of a route. In
both systems, the authors themselves collected the necessary photographs. There is
no automatic identification of potential landmarks. Also, given the time of their
development (late 1990s and early 2000s, respectively), both approaches run as
desktop web browser application, and are clearly outdated in their technology.
Only a few years later, the first mobile applications using photograph-based
navigation appeared (e.g., [ 32 , 44 ] ). Beeharee and Steed [ 3 , 4 ] used photos to
augment wayfinding instructions presented on mobile devices (here, personal digital
assistants, or PDAs for short). While this clearly made the system more useful,
since now users can directly compare what they see in the environment with what
they see on the photograph, photographs are still selected by the authors. There
is no indication how this system might scale up, even though, the authors state
that technically it would work anywhere (in the UK). You can clearly observe
the transition to today's modern smartphone applications, but a lot remains rather
awkward in this early system, though in some sense it was ahead of its time.
Hile et al. [ 34 ] presented another photograph-based navigation service for mobile
phones. The system uses a database of previously geotagged photographs from
an outdoor augmented reality project [ 60 ] , which contains photographs of mostly
the Stanford University campus. The system segments an environment into loxels,
which are 30 30 m grid cells. For each loxel the current route runs through, the
system determines all photographs visible from that loxel. Based on nearness to the
route and deviation from movement direction a cluster of photographs gets selected
with each cluster containing a set of photographs with similar views. The center
of the cluster provides a canonical view, similar to those approaches discussed
in Sect. 5.2.2 . This canonical view is presented to the users along with textual
instructions on how to proceed along the route. Additionally, an arrow is overlaid
on the photograph indicating the direction to take.
Similar to the system of Beeharee and Steed, the system of Hile et al. faces the
issue of how to provide a sufficient number of useful photographs. This links back
to our discussion about crowdsourcing as an alternative approach to data collection
(see Sect. 5.5.3 ) .
Recently, Microsoft Research presented a pedestrian guidance system using
street level panorama images [ 68 ] . The system is based on Nokia City Scene , 5 which
combines panoramic photographs, city models constructed from LIDAR data, and
map data of about 20 of the major cities around the world. City Scene is similar to
Google Street View, but with significantly less coverage. It seems to focus mainly
on the USA, and at the time of writing it is restricted to the Nokia N9 smartphone.
The system highlights specific features as landmarks, primarily business signs. In
case no signs are available, a building is picked instead. A text recognition pipeline
is able to automatically extract signs from the photographs determining for the
5 http://store.ovi.com/content/178183 , last visited 8/1/2014.