Improving Geolocation by Combining GPS with Image Analysis - Geoinformatics for Intelligent Transportation - page 207

Keywords Computer vision Geolocation GPS A-GPS Image analysis

Pattern recognition

1 Introduction

In recent years, the means used in the calculation of geographic location have
evolved, becoming progressively more accurate. Some of the methods and tech-
nologies used in geolocation can give an accurate location on the Earth ' s surface,

but not an exact location [ 1 ]. In the field of geographic location, currently, there are

three main technologies: Global Positioning System (usually known as GPS),
Assisted-GPS (usually known as A-GPS) and Cell tower ID. The

first, GPS, is
based on a set of geostationary satellites and a computation having as input the GPS
signal from those satellite and as output a location on the Earth s surface. However,

GPS has two main drawbacks: its signal is highly affected by noise and it requires
direct vision between the GPS receiver and a set of at least four satellites [ 2 ]. To
compensate this problems, and accelerate the positioning, the A-GPS technology
was developed, combining GPS information with information from the network. As
the main characteristic of this technology indicates, it is internet dependant, which
means that the major strength of A-GPS does not work everywhere [ 3 ]. Finally, the
Cell Tower ID, is a GPS-free technology, that uses only cellular network to ref-
erence a position. This technology uses cell coverage to determine the position of
some device, but it is not much accurate [ 4 ].
The geolocation technologies above presented perform well for most of the
cases. However, for some situations, GPS and A-GPS may not be able to provide
geographic location. For instance, when a vehicle suddenly enters a long urban
street cutting through skyscrapers, the GPS signal may be affected by the metal
structures of the buildings and the lack of direct vision from the receiver to the
satellites.
Our approach tries to overcome this problem by combining computer vision
(CV) with GPS, regarding the scope of geolocation, where CV is used to help in the
geolocation process when GPS signal is weak or not available.
Computer vision has evolved in the last decade and its applications are becoming
more comprehensive. As Bernal stated [ 5 ], when we think about computer vision, it
is impossible not to think about using features. In CV, many feature descriptors
have already been proposed and tested. According to Bernal, feature descriptors can
be divided into four main groups: texture descriptors, colour descriptors, shape
descriptors and motion descriptors [ 5 ]. For the purpose of the current work, a study
of several descriptors was previously performed and three main descriptors were
selected as being most promising: Scale Invariant Feature Transform (SIFT, [ 6 ]),
Speeded-up Robust Features (SURF, [ 7 ]) and Histogram of oriented gradients
(HOG, [ 8 ]). The three algorithms present good results, in terms of analysis success
rate and response time, but the HOG was slightly inferior to SIFT and SURF.

'