Humans can easily categorize behavior, making quick observations about what's normal and not. Now, computers are catching up. A new breed of software can keep track of—and understand—the behavior of people, animals and machines. What's more, it can respond appropriately to the behavior it observes in order to increase security and productivity. It's even helping in the care for the elderly.

One example of this "behaviometric" software is an application written by two students at the Technion-Israel Institute of Technology in Israel. Their application can identify people with close to 100% accuracy by analyzing their typing styles. In particular, the software records the number of milliseconds it takes for a typist to press and let go of a key. And because typing a particular letter has an effect on how the next key is pressed and the number of key combinations is finite, the application can figure out the probability that the individual doing the typing is the "target" person.

Another behavior-detecting software is made by Atlanta-based VistaScape Security Systems. Its application analyzes video surveillance tapes for such high-security facilities as shipping ports, airports and petrochemical plants. Utilizing general image templates of people, objects, animals, etc, the software compares activity against rules supplied by the client. For example, some rules could prohibit people from entering certain areas or scaling barricades.

A whole new set of rules, meanwhile, comes into play in a system developed by New York City-based Living Independently. Its software monitors elderly people living at home, notifying family or health professionals of any sudden changes of behavior. "Working with gerontologists, we identified the behaviors that would be most valuable to monitor," says George Boyajian, executive vice president for strategy, research and development. These include rising from bed, approaching a food preparation area, opening a medicine cabinet and going into the bathroom. Base stations pick up such activities with infrared sensors and transmit the data over a regular phone line to the company. Proprietary algorithms match the behavior against what it regards as "normal" for that person based on previously gathered information. The system then alerts family members or health care professionals if behaviors suddenly increase in frequency or are abruptly reduced or discontinued. Because "typical" behavior can change over time, the software periodically adjusts its baseline of behavior.

Indeed, for behaviometric software to be effective, machines must be able to learn and adapt. While humans can easily distinguish between myriads of object behaviors—from birds taking flight to boats gliding on the surface of the water—based on their experience, computers have to work hard to understand these activities. The single most formidable obstacle for these applications is sorting through huge amounts of information. Fortunately, researchers in behavior-monitoring software are finding ways to boil down a behavior to its simplest and most essential form while still giving the application enough statistical information to categorize an action and make a sound decision.

Source:

Behavior-Monitoring Machines
Erik Sherman
Technology Review, November 12, 2003
www.technologyreview.com/articles/wo_sherman111203.asp