After more than 50 years of research, machine vision technology has recently reached the point in its development where it is ready for commercial application. Now, the rush is on to make the most profitable use of it.
Machine vision technology teams up specialized computer hardware, cameras, and software to create a machine that can see and recognize objects and people and then make useful interpretations of those images. One application described recently by CNet News,1 is a system from Hyperactive Technologies that monitors operations at fast food restaurants. That machine vision system keeps track of cars coming into the parking lot and then compares the anticipated demand to the current amount of cooked food that is available. By calculating cooking times, it can alert the staff ahead of time as to what food items they will need to prepare next, thereby improving efficiency, as well as customer satisfaction. The system is currently being tested at Popeye's Chicken and Jack in the Box outlets. Zaxby's, which already has the system installed at more than 100 restaurants in its chain, says it is saving over $800,000 a year because the system wastes less food.
According to The Economist Technology Quarterly,2 these types of systems are being used to improve efficiency in the manufacture of everyday goods, such as diapers. If a more accurate machine vision system can reduce the wasted material that is cut by even just one millimeter per diaper, it can save millions of dollars for a company that produces billions of diapers.
Another example involves companies that sell consumer packaged foods, such as rice. Rice sorters are using computer vision to scan four tons of rice an hour and then use air jets to eject discolored grains, rocks, or other debris. That requires the extremely fast and accurate computer vision systems that are now becoming widely available.
Omron Corporation in Japan is developing a system that can tell if employees, such as cashiers, are smiling at customers or not. The same sort of system can be used to determine if an expression is authentic or not. Unilever is using it to determine how food tasters are reacting to a product. Procter & Gamble is also using it to monitor responses of participants in focus groups to see if their words match their facial expressions.
MICTA, a media laboratory in Australia, is testing ways to use machine vision to target advertising. A digital billboard called TABANAR, with a camera installed in it, reads a person's gender, age, and hair growth. It can then decide which of several ads to display. For example, it might present an ad for action figures to little boys or an ad for razors to men who shave. If the system sees that the person is ignoring that ad, the system can switch to a potentially more attractive ad.
On the other hand, cigarette machines in Japan are now equipped with a machine vision system that estimates a person's age by measuring facial fat. It can switch off the machine if the person appears to be too young to buy tobacco. The system has proved to be more accurate than bouncers in nightclubs at estimating age.
The Toronto Rehabilitation Institute in Canada is testing a system for monitoring elderly people in nursing homes or when they are alone. If a person falls or stops moving, it can alert someone. It can also remind the seniors to brush their teeth.
Similar software can monitor workers, identifying those who take long breaks or don't work quickly enough. Workers in automobile factories can be alerted if components are missing or not fitting correctly. They can even be warned if they reach for the wrong tool or part.
Autonomy, a company in England, has a machine vision system that can read a car's license plate, and then see if it matches the make, model, and color of the car that's supposed to have that plate. When there is a mismatch — as there might be on a stolen car — the system alerts the police. A similar system that is being tested in Istanbul resulted in the arrest of 15 drivers in two months.
Another vision system for automobiles, called Mobileye Vision Technologies, is being tested in Israel. It helps drivers to identify blind spots and watches out for possible traffic conflicts. BMW, General Motors, and Volvo have already bought the system, which will soon be capable of issuing collision warnings and even putting on the brakes.
DFS Deutsche Flugsicherung, a government agency in Germany, will soon introduce a machine vision surveillance system for airports to avoid conflicts among taxiing aircraft and other vehicles.
The consulting firm Accenture is about to launch its Accenture Mobile Object-Recognition Platform, a free service that allows people to send images via mobile phones. The images are processed and identified in much the same way that search terms are used for finding things on the Internet. For example, you could snap a photo of a product, and the system would then return the locations of places to buy it. But the uses of such a system are seemingly infinite. Microsoft is developing a similar system called Lincoln, which is already able to identify over a million objects.
Meanwhile, a company called Cognex is developing computer vision systems for controlling and monitoring robots in factories and other settings. The robots themselves are being equipped with cameras at the University of British Columbia. They then learn through a machine vision system to recognize objects. Eventually, the robots will learn to use the objects, as well as avoid bumping into them.
In light of this trend, we offer the following five forecasts for your consideration:
First, object recognition technology has finally matured to the point where practical applications will snowball. The scramble is on to find the winning consumer applications for it. A Google search for "machine vision" returns more than 15 million hits. Look for companies like Microsoft and Google to pursue this field aggressively, even as scores of start-ups attempt to hit on the killer app that will capture the hearts and minds of users everywhere.
Second, a wide range of commercial and consumer computer vision applications will emerge simultaneously and benefit from each other. While some companies strive to put machine vision into the hands of consumers, others are focusing on applications for businesses and corporations. Expect to see a diverse range of tailored products aimed at helping companies increase their efficiency with very specific processes. For example, in the medical field, machine vision systems are in use to examine hypodermic needles to make sure that they are sharp and free of defects. The number of industrial processes that can be improved through machine vision is limitless, and companies that specialize in customizing this technology will reap huge rewards by saving money for businesses and customers.
Third, in the beginning, machine vision will confer competitive advantage on early adopters. But as the technology spreads, it will simply become a cost of doing business. In some industries, it will replace people, requiring a shift in training for many corporations. Those companies who retrain their people in the use of these new technologies will be ahead of the game as they become the leaders in making the most of them.
Fourth, a huge untapped market exists in the military, in security, and in law enforcement. Identifying license plates is just the tip of the iceberg. Security cameras are already ubiquitous. As they begin to be hooked up to machine vision systems, they can serve to fight crime in numerous ways, protecting premises from unauthorized people and identifying criminals who are at large. The military is already experimenting with systems that identify suspicious vehicles in places like Iraq. Companies that focus on these applications of machine vision technology will present excellent entrepreneurial and investment opportunities.
Fifth, machine vision will accelerate the evolution of the Semantic Web. As people are able to communicate with the Internet via photos and video instead of having to describe things in words, machine vision will make the content of the Web much more useful. As machine vision becomes more adept at recognizing objects, and as objects are identified with RFID tags, the true dream of the Semantic Web will come true, probably beginning within the next five years.
| References |
|---|
|