Ms. Negrin, who has coloboma, a birth defect that perforates a structure of the eye and afflicts about 1 in 10,000 people, is an employee at OrCam, an Israeli start-up that has developed a camera-based system intended to give the visually impaired the ability to both “read” easily and move freely.
Until now reading aids for the visually impaired and the blind have been cumbersome devices that recognize text in restricted environments, or, more recently, have been software applications on smartphones that have limited capabilities.
In contrast, the OrCam device is a small camera worn in the style of Google Glass, connected by a thin cable to a portable computer designed to fit in the wearer’s pocket. The system clips on to the wearer’s glasses with a small magnet and uses a bone-conduction speaker to offer clear speech as it reads aloud the words or object pointed to by the user.
The system is designed to both recognize and speak “text in the wild,” a term used to describe newspaper articles as well as bus numbers, and objects as diverse as landmarks, traffic lights and the faces of friends.
It currently recognizes English-language text and beginning this week will be sold through the company’s Web site for $2,500, about the cost of a midrange hearing aid. It is the only product, so far, of the privately held company, which is part of the high-tech boom in Israel.
The device is quite different from other technology that has been developed to give some vision to people who are blind, like the artificial retina system called Argus II, made by Second Sight Medical Products. That system, which was approved by the Food and Drug Administration in February, allows visual signals to bypass a damaged retina and be transmitted to the brain.
The OrCam device is also drastically different from Google Glass, which also offers the wearer a camera but is designed for people with normal vision and has limited visual recognition and local computing power.
OrCam was founded several years ago by Amnon Shashua, a well-known researcher who is a computer science professor at Hebrew University here. It is based on computer vision algorithms that he has pioneered with another faculty member, Shai Shalev-Shwartz, and one of his former graduate students, Yonatan Wexler.
“What is remarkable is that the device learns from the user to recognize a new product,” said Tomaso Poggio, a computer scientist at M.I.T. who is a computer vision expert and with whom Dr. Shashua studied as a graduate student. “This is more complex than it appears, and, as an expert, I find it really impressive.”
The advance is the result of both rapidly improving computing processing power that can now be carried comfortably in a wearer’s pocket and the computer vision algorithm developed by the scientists.
On a broader technology level, the OrCam system is representative of a wide range of rapid improvements being made in the field of artificial intelligence, in particular with vision systems for manufacturing as well as fields like autonomous motor vehicles. (Dr. Shashua previously founded Mobileye, a corporation that supplies camera technology to the automobile industry that can recognize objects like pedestrians and bicyclists and can keep a car in a lane on a freeway.)
Speech recognition is now routinely used by tens of millions of people on both iPhones and Android smartphones. Moreover, natural language processing is making it possible for computer systems to “read” documents, which is having a significant impact in the legal field, among others.
There are now at least six competing approaches in the field of computer vision. For example, researchers at Google and elsewhere have begun using what are known as “deep learning” techniques that attempt to mimic biological vision systems. However, they require vast computing resources for accurate recognition.
In contrast, the OrCam technique, which was described in a technical paper in 2011 by the Hebrew University researchers, offers a reasonable trade-off between recognition accuracy and speed. The technique, known as Shareboost, is distinguished by the fact that as the number of objects it needs to recognize grows, the system minimizes the amount of additional computer power required.
“The challenges are huge,” said Dr. Wexler, a co-author of the paper and vice president of research and development at OrCam. “People who have low vision will continue to have low vision, but we want to harness computer science to help them.”
Additionally the OrCam system is designed to have a minimal control system, or user interface. To recognize an object or text, the wearer simply points at it with his or her finger, and the device then interprets the scene.
The system recognizes a pre-stored set of objects and allows the user to add to its library — for example, text on a label or billboard, or a stop light or street sign — by simply waving his or her hand, or the object, in the camera’s field of view.
One of the key challenges, Dr. Shashua said, was allowing quick optical character recognition in a variety of lighting conditions as well as on flexible surfaces.
“The professional optical character readers today will work very well when the image is good, but we have additional challenges — we must read text on flexible surfaces like a hand-held newspaper,” he said.
Although the system is usable by the blind, OrCam is initially planning to sell the device to people in the United States who are visually impaired, which means that their vision cannot be adequately corrected with glasses.
In the United States, 21.2 million people over the age of 18 have some kind of visual impairment, including age-related conditions, diseases and birth defects, according to the 2011 National Health Survey by the U.S. National Center for Health Statistics. OrCam said that worldwide there were 342 million adults with significant visual impairment, and that 52 million of them had middle-class incomes.
(Source: nytimes.com)
No comments:
Post a Comment