The next step in digital video?
Vision Of Tomorrow -- Seeing Is Computing
Aug. 10, 1998 (Computer Reseller News - CMP via COMTEX) -- Thanks to
voice-recognition software, computers are beginning to talk and listen.
But they have, to date, remained primarily blind.
That, however, is going to change as researchers improve the way
computers recognize image patterns. In fact, analysts and researchers
said that within the next five to seven years, personal computers will
add another sense-the ability to "see."
There are several ways computers will first start to put their
newfound abilities to use, said Aaron Bobick, associate professor of
computational vision at the MIT Media Laboratory, Cambridge, Mass. Some
will be based on computer vision capabilities of today, which already
are starting to play a role in controlled industrial environments and
high-end modeling for medical applications, he said.
A large part of the problem in getting computers to see is not how
much computational power is available, said Matthew Turk of Redmond,
Wash.-based Microsoft Corp.'s Microsoft Research arm. Rather, it is
knowing how to frame the problem of "seeing" in a way the computer can
understand and process, so data flows through its camera, or eyes, Turk
said.
"There has been a whole vision community out there for 30 years
scratching their heads wondering how to give computers vision. We are
certainly making progress, but we're not quite there yet," he said.
"The biggest hurdle is knowing how to construct the problem. It's not a
bandwidth or computational problem. Sometimes when you get faster
computers you just get to scratch your head faster."
Still, even though computers will not soon be able to interpret
classical art or the ballet, they already are getting pretty good at
recognizing football plays, said industry observers.
"If the New England Patriots run a play, the system can actually
start to tell what play they ran," said MIT's Bobick.
Trying to get the computer to interpret a raised eyebrow or one of
the other thousand of small and subtle facial movements people make,
however, is a different ball game.
"Many people wonder why, if we can get the computer to read football
plays, we can't get the system to read facial expressions. Well, there
are an awful lot of things that have to happen in each given play, and
there are a very small number of plays, maybe 100," said Bobick.
"This gives us a lot of evidence for each play so we can build local
relationships between players. Such as, if the center hands off to the
quarterback and there is a certain block, and through seeing this
information the computer is capable of deducting reliable answers as to
what play just occurred," he added.
In fact, entertainment could be the first area to benefit from
computers' upcoming improved visual capabilities, said research
experts.
For example, Media Lab conducted KidsRoom, a project where the
researcher used computer vision as input for the narration of an
interactive story. As children moved throughout a room, the story
reacted in realtime based on what it "saw."
"These kinds of things allow you to have real interactive experiences
in a large scale. So instead of point and click on a screen, it's a
room-size scale and here computer vision will provide an opportunity
for entertainment. At first, it's going to be at the large-scale
entertainment centers built by the likes of Universal Studios, Disney
and Sony; places willing to put in the effort and time needed to build
something no one has done before," said Bobick.
Another up-and-coming market is security.
Although systems will be unable to tell if something wrong or
criminal is happening at a mall, for example, they will be able to look
at crowded areas and tell if things do not seem normal, experts said.
Systems would then notify a security department that could send a guard
to investigate. Vendors and VARs will build vision-equipped systems for
banks, automated-teller machine environments and the government to help
reduce some types of crime, said both Bobick and Turk. For example, an
ATM could be equipped to see that the person who entered the ATM card
is not the rightful account holder, experts said.
Computers are still a long way from meeting the challenge of reading
expressions and facial movements. But organizations are making
progress, research executives said.
"We are working on the goal of being able to see where a user is
looking on the screen, so the computer can prepare in advance, even a
couple of milliseconds," said Turk.
In the future, the graphical user interface could see that users are
about to print when they look at the printer icon for a certain length
of time, he said.
Integrators can see early inklings of this bleeding-edge research
today. Microsoft Research recently made available version 1.0 of its
Vision Software Development Kit, a low-level language library of object
definitions and related software for use with Visual C++ to help
developers and researchers create image-processing and acquisition of
live elements.
George V. Hulme is a freelance business and technology writer based
in Croydon, Pa.
-0-
By: George V. Hulme
Copyright 1998 CMP Media Inc.
|
