Geography Reference
In-Depth Information
Keywords Spatio-temporal clustering
Semantic enrichment
Geotagged photo
collections
1 Motivation
Ubiquity of location-aware devices, cheap storage and fast computing power has
enabled collection and analysis of large amounts of spatio-temporal data. Different
application domains like zoology, activity-based analysis or tourism in which data
collection was a tedious and manual process (observation, surveys), benefit from
the positioning technology and demand new analysis and techniques to cope with
large quantities of these data.
Collections of geotagged photos have recently become available (Goodchild
2007
) due to the availability of photo-sharing sites such as Flickr (
http://
www.flickr.com
.
com) and Panoramio (
http://www.panoramio.com
)
, in which
millions of users from all over the world upload their geo-referenced photos. The
basic information provided by a person during photo upload is the location where
the photo was taken, the time of the action, and the textual identifiers including
title and tags. The photo may also be a member of some thematic group. A photo
taken by a person can be regarded as an event, and collection of photos of a person
can be considered as a trajectory. Such user-generated data have already been used
in the analysis of attractive places (Crandall et al.
2009
; Kisilevich et al.
2010a
),
movement behavior (Girardin et al.
2008a
) and mobility (Andrienko et al.
2009
).
The advantages of these data are (Girardin et al.
2008b
): unlike the automatic
capturing of traces, the manual disclosure of location in the act of geotagging of
photo provides additional qualities: positioning a photo on a map is not simply
adding information about its location; it is also an act of communication which
contains what people consider as relevant for themselves and others.
Until now, these data were used as an alternative to the GPS-based data, mainly
utilizing coordinates and timestamps. However, the title, tags, thematic group
name as well as the photo itself, may reveal the context of the photo or describe the
place where it was taken: some known event, a landmark or a person. Multimedia,
computer vision and text mining communities realized the potential of geotagged
data (Toyama et al.
2003
) and proposed automatic approaches for such tasks as
image summarization (Kennedy et al.
2007
; Zheng et al.
2009a
), landmark iden-
tification (Crandall et al.
2009
), automatic event identification (Kennedy et al.
2007
; Ahern et al.
2009
; Becker et al.
2009
), which includes clustering and
retrieval of tag representatives. Information retrieval methods allowed automatic
gazetteer creation using geotagged images (Popescu et al.
2008
), Wikipedia, and
web search engines and ontology induction from tagged images (Schmitz
2006
).
However, pure automatic approaches of event or place exploration have several
disadvantages that are important to draw attention to.
