Select your localized edition:

Close ×

More Ways to Connect

Discover one of our 28 local entrepreneurial communities »

Be the first to know as we launch in new countries and markets around the globe.

Interested in bringing MIT Technology Review to your local market?

MIT Technology ReviewMIT Technology Review - logo


Unsupported browser: Your browser does not meet modern web standards. See how it scores »

{ action.text }

Researchers in behavior-detecting software find that they must walk a tight rope: seeking to represent the essentials of behavior a problem in the simplest possible way that still carries enough statistical information to allow the computer to place an action into a category and make an appropriate decision. And the single biggest obstacle is having too much information. Data filtering-passing on only the cats and dogs and tossing out rocks and trees-is vital. Without it, the fastest computer would grind to a halt under the weight of detail.

“We look at it as a data reduction problem,” says Glenn McGonnigle, CEO of VistaScape Security Systems, an Atlanta-based company makes software that examines video surveillance tapes for such high -risk facilities as shipping ports, airports, and petrochemical plants. Using generic image  templates of objects-people, boats, birds, and so forth-VistaScape’s software matches activity against rules provided by client. One rule, for instance, might forbid people to climb a fence and enter a sensitive area.

To alert security personnel to potential problem situations, the software must sift through a bewilderingly large number of object behaviors: water undulating in waves, boats moving against the horizon, birds darting and soaring through the air, people strolling on a beach. From the cradle, people learn to instantly make these distinctions, using experience to increase their powers of discernment; machines lack that advantage.

But whether something is important or superfluous depends completely on what the scientists are trying to detect. For McGonnigle’s purposes, vital details might include object size and speed, while color might be irrelevant. Direction of motion would matter in some cases-someone walking one way through an airline security check would be normal, while travel in the opposite direction could indicate a potential problem. Delivery trucks arriving on a Tuesday or Thursday might be expected, though a Wednesday appearance might warrant an investigation.

Since what is important in one context is of no interest in another, each behaviometric system enlists subject experts to help define what behavior will be important. New York City-based Living Independently recently began shipping a system that keeps tabs on elderly people living at home, alerting family or health professionals if behavior suddenly changes. Living Independently had to know what actions might shine some light on the state of someone’s health. “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.

As it turned out, the list of considerations could be pared to a bare minimum. Infrared sensors note when someone gets up, nears a food preparation area, opens a medicine cabinet, and enters the bathroom. Data travels wirelessly to a base station, which relays the information over an ordinary phone line to the company. Proprietary algorithms compare the actions to what the system considers “normal” for the person based on previously collected data. Sudden changes in behavioral patterns-more frequent night time trips to the bathroom, for example, or a sudden cessation of opening the medicine cabinet to obtain pills-can be early signs of trouble, or might indicate an immediate problem. The system then sends alerts to family members or appropriate healthcare professionals.

Normal is, of course, relative, and can change over time. That is why behaviometric software must also incorporate machine learning, so that decisions are not made based on some behavioral standard that is no longer applicable. Living Independently creates a baseline of behavior that changes over time. Each person is observed, with software abstracting statistically “typical” behavior on a roughly ten day rolling time frame, shifting as someone’s habits change. Thus, the software avoids raising alarms unnecessarily.

Behaviometric software is not perfect. The Living Independently application, for example, can only tell if people approach food preparation areas-not if they eat. A ship nearing an offshore oil rig might be only a lost pleasure craft, and not terrorists planning an assault. Still, an approximation is sometimes all that’s needed to answer a question. So the name time you are wondering how Aunt Sadie is doing-or who is at her keyboard-ask her computer. It might just tell you.

0 comments about this story. Start the discussion »

Tagged: Computing

Reprints and Permissions | Send feedback to the editor

From the Archives


Introducing MIT Technology Review Insider.

Already a Magazine subscriber?

You're automatically an Insider. It's easy to activate or upgrade your account.

Activate Your Account

Become an Insider

It's the new way to subscribe. Get even more of the tech news, research, and discoveries you crave.

Sign Up

Learn More

Find out why MIT Technology Review Insider is for you and explore your options.

Show Me