Chip makers join embedded vision group

May 31, 2011 // By Rick Merritt
More than a dozen vendors have joined a new industry group promoting computer vision in embedded systems. The Embedded Vision Alliance aims to provide education about the technology and help develop a broad range of emerging markets for it.

Founding members include Analog Devices, Avnet Electronics Marketing, Ceva, Freescale, MathWorks, National Instruments, Nvidia and Texas Instruments.

"I personally believe this technology is going to have a huge impact making all sorts of system more intelligent, aware and responsive," said Jeff Bier, president of Berkeley Design Technology Inc. (BDTI), a DSP consulting firm and organizing member of the group.

The group shares a belief that computer vision technology is maturing, making it ripe for use in everything from toys to TVs, car navigation systems, medical systems and more. Bier shared that vision in a talk in April when he first tipped plans for the group.

"Just look at the Microsoft Kinect, which added vision to the Xbox 360 and became the fastest-selling consumer electronics device in history, shipping more than 10 million units in 5 months—and that's just a small part of the story," said Bier.

The group aims to provide education about the technology and its markets through its Web site. However, members are also considering technical roles the group could play.

"There is a strong need for standards, and there also is a need for a tangible, common infrastructure for the embedded developer," said Bier, pointing to OpenCV, a PC-based tool. "That’s great, but what about an inexpensive, easy-to-use embedded vision platform," he asked.

Plenty of technical challenges are still ahead in embedded computer vision, according to Bier.

The technology can demand significant silicon performance. For example, running an algorithm on a 720-progressive flow of imaging data optimized for a modern VLIW DSP architecture can consume about 200 MHz/frame/second, he said. In addition, many vision functions will require highly parallel or specialized hardware and the algorithms they run are diverse and dynamic, he added.

Dues for a basic membership are only, $5,000, a factor that is attracting a number of smaller companies. They include Apical Imaging, a U.K. company developing algorithms to improve image quality; Ximea, a company