Magazine

The Vision Thing Gets Closer for Computers


Giving devices the power to comprehend the three-dimensional world around them has long been a central pursuit of machine-vision researchers. Cheap digital cameras have made it simple enough to generate images that computers can process. But true machine vision requires that a device be able to make sense of what it sees, a much harder and more expensive task.

By keeping things small and simple, San Jose (Calif.) startup Canesta may be close to providing a sort of primitive 3-D vision to a variety of inexpensive products. Called "electronic perception," the approach uses infrared light and a sensor chip like those in digital cameras. Light is beamed at the target, and the tiny differences in the time it takes rays to return to specific points on the chip are measured. Gauging distance based on the time that signals travel to and from a target is the principle behind radar. Canesta uses the time differences to compute a three-dimensional relief map of the target.

Canesta's patented sensor can do 50 scans per second and resolve features as small as 1 millimeter. Company CEO Nazim Kareemi says volume production of the sensors would allow the required software and hardware to be incorporated into devices for as little as $10. Enabling stereoscopic vision for such a small sum of money opens up all sorts of possibilities. Instead of a conventional controller, for example, a game console could translate the motions of a player's body into movement on the screen. If all goes well, the first products using the Canesta technology could hit the market by the end of this year, says Kareemi. Last December, IBM (IBM) boasted it was building the world's fastest weather supercomputer, called Blue Storm, for the European Center for Medium-Range Weather Forecasts. But even before Blue Storm's startup late this year, Big Blue has lost bragging rights. NEC Corp.'s (NEC) Earth Simulator system was switched on in March, and it is twice as fast.

Installed at the Yokohama Institute for Earth Sciences in Japan, NEC's supercomputer harnesses 5,120 processors to achieve an incredible peak speed of 40 trillion calculations a second, or 40 teraflops. Blue Storm's top speed will be 20 trillion.

Researchers in Yokohama will use Earth Simulator to run the highest-resolution model of the earth yet created. The model wraps the entire planet in a virtual grid of small boxes. To simulate weather and climate patterns, predict pollution levels in the oceans and atmosphere, and study other geophysical trends, the computer solves equations for each box, then combines them to derive a worldwide perspective--compressing centuries of global change into three days of computing, not the months it now takes. Being able to run multiple variations of a problem in rapid succession, researchers predict, will lead to new insights about global environmental trends. Two groups of scientists have succeeded in mapping the rice genome in an important breakthrough that could help ease world hunger. Because rice is a dietary staple for more than half the world's people, scientists are hopeful the information gleaned from its genome will lead to improvements in nutrition, crop yield, and sustainable agriculture.

The two reports, in the Apr. 5 issue of Science, mark the first time a crop plant has been sequenced. A team led by scientists from the Beijing Genomics Institute deciphered the genetic code for indica rice, a major crop in the Asia-Pacific region, while researchers from Myriad Genetics in Salt Lake City mapped the japonica strain, favored in arid regions. Both strains are incredibly complex, with 45,000 to 56,000 genes in indica and 42,000 to 63,000 in japonica. By comparison, the human genome has only 30,000 to 40,000 genes--although human genes are 16 to 36 times longer. The scientists believe that differences in gene length may indicate different mechanisms for generating diversity. Emergency room doctors treating patients in life-threatening situations have a powerful new diagnostic weapon available to them. A medical imaging technology recently unveiled by Mercury Computer Systems can produce a 3-D X-ray image in about 15 seconds. Using ordinary X-ray gear, it takes doctors 5 to 10 minutes to create such an image.

Mercury's secret is a dedicated circuit board that can run image-reconstruction algorithms far more quickly than a PC microprocessor. It would take 15 processors working in parallel to achieve comparable speeds.

Outside the ER, Mercury's technology could aid doctors performing common procedures such as angioplasty. Currently, a physician inserting a catheter into a coronary artery to clear a blockage is guided by two-dimensional X-ray images. To get a different view, the doctor must reposition the patient and take another X-ray. The new technique lets the doctor work and view 3-D images in almost real-time. "It was beyond our expectations that this could be done," says Michael W. Vannier, a radiologist at the University of Iowa.


Tim Cook's Reboot
LIMITED-TIME OFFER SUBSCRIBE NOW
 
blog comments powered by Disqus