Aug. 24, 2006 — A wearable computer system that emits audio cues could help guide the visually impaired.
The System for Wearable Audio Navigation, or SWAN, combines global
positioning system (GPS) technology with cameras and image processing
software to locate a person's whereabouts and "see" details such as windows, doors and corners.
"In the future, we could even use the cameras to recognize people or
objects," said Frank Dellaert, who developed the
system at the Georgia Institute of Technology with Bruce
Walker.
The technology could do everything from lead a blind person through a
new neighborhood to help firefighters or soldiers plot a course in darkness.
Since about 2001, visually impaired people have had access to
commercial GPS-based navigation systems meant to help them get around.
The
technology is available on handheld electronic devices such as laptops
or PDAs and is similar to that used in cars. A GPS sensor pinpoints the person on a grid, while the computer's
database — which contains street names as well as businesses — matches the information to the location.
Menu choices and directional cues are provided either through a
Braille-based readout or a synthesized voice, which can be heard
through a speaker or earphones.
But GPS-based systems have some disadvantages, said Dellaert. For one thing, the
signal may not come through if a person is walking through a city
dense with tall buildings. If the person stops walking but turns, GPS can't tell what direction the person is facing.
Georgia Tech's solution was to incorporate additional sensors on their
navigational system — including a digital compass, four cameras, a tilt
sensor and a light sensor.
The compass, GPS, and tilt sensors track the direction the user is walking in, or if still, facing.
The cameras image the surroundings, while the light sensor measures
the brightness of objects. Computer vision software interprets the
scene based on the configuration of shapes caused by
features such as corners, edges, light and dark areas, and repetitive
patterns.
For example, a particular arrangement of edges, dark space and
corners could correspond to a door. The configuration could also be
matched with a database of known features to determine what's nearby.
