Sherri Barber/The Coloradoan
NEW TECHNOLOGY: Monica Chawathe, a doctoral candidate at Colorado State University,
holds a computer card Thursday. Chawathe is a part of the team that helped
develop a new computer software technology that improves a computer's speed
in processing visual images.
The team of Colorado State University computer scientists couldn't quit smiling.
In the palm of team member Monica Chawathe's
hand was a computer board, about the size of credit card, the fruit of a
four-year project that likely will change the way visual images are transmitted
Written onto the hardware, called a Field
Programmable Gate Array, or FPGA, is the team's compiling program, which
works better and faster than they thought it would.
In the lightning-fast world of the Pentium processor, their programmed FPGA is 600 times faster.
"It's a very different compiler; it works
at drastic speed," said Charlie Ross, a CSU doctoral candidate and team member.
"It's not doing things (other compilers do) the same way. It's a totally
The software will have a vast number of uses to aid in the recognition of visual images, said team leader Wim Bohm.
The project was funded by Defense Advanced Research Projects Agency, or DARPA.
The program will be used by the U.S. military
in night-vision target recognition systems to identify objects such as enemy
tanks. The Air Force plans to use the program to scan satellite photos to
distinguish dot-sized vehicles and facilities from otherwise random dots
and do it fast.
"If you are in a military operation in
the middle of the night trying to determine if an enemy tank is in the vicinity,
you don't want the answer in an hour or even a minute, you want it right
now," said team member Bruce Draper. "A system with an 800 MHz Pentium chip
displays the image in 65 seconds: our system takes less than one second."
The program spits out the answer faster
than the question can be asked, Ross said. Although it might first be used
by the military, its uses could be vast, he said.
The team of three CSU professors -- Bohm,
Draper and Ross Beveridge -- and a rotating group of eight graduate students
spent four years writing the compiler and then just two hours to write the
first detection program to see if the compiler would work.
Think of the compiler as sort of putting
an architect on a chip. A specific program might ask the architect to design
a building with "x" number of floors, windows, doors, bathrooms and square
feet, and the architect chip would process the information and spit out a
floor plan. The compiler works much the same way, Ross said.
Previous programs would have to create
the architect and decipher the information at the same time. In this program,
the architect is already there.
The team's compiler worked on the first try, even better than expected, Bohm said.
DARPA got more than CSU promised, he said.
In applying for the grant, CSU described the best-case scenario. But this
case was even better, he said.
The compiler has possible uses in robotics,
night-vision recognition for the blind and security systems that use face-recognition.
Neuroscientists are even curious to see
if a chip using the compiler can be implanted into the brain of a person
who has seizures to detect the "brainstorms" before they happen, Bohm said.
If it works, the patient could take medication
to avert seizures rather than taking medication all the time to prevent them,
Such futuristic applications are in the what-if stage for now, he said, but it's exciting.