As we age, many of us want to live in our own home for as long as possible. In the United States, for example, this desire is shared by more than 85 percent of people older than 65, according to a study by AARP.
When chronic health issues make it harder to live independently, however, many senior citizens eventually enter assisted-living facilities or nursing homes, or move in with relatives. In an effort to extend the time that people can stay in their own home, IEEE Fellow Diane Cook is helping to design a smart-home system with motion-tracking sensors to monitor seniors’ activities as they go about their daily routines.
Cook, a professor of electrical engineering and computer science at Washington State University in Pullman, is director of the university’s Artificial Intelligence Laboratory and its Center for Advanced Studies in Adaptive Systems (CASAS).
She and her research group last year received a five-year, US $1.8 million grant from the U.S. National Institute of Nursing Research to continue their “clinician in the loop” project. The program collaborates with local assisted-living facilities.
HOW IT WORKS
The group’s approach is keyed to the needs of each individual. First, a health care professional identifies data that’s relevant to the person’s health and safety. From that information, engineers create algorithms to recognize behavioral patterns related to physical or cognitive health. Cook estimates that her team at CASAS has installed about 140 smart-home systems as part of a pilot program to test the system.
Its hardware is simple. Passive infrared motion sensors installed around the house detect where people are, based on their body heat, and note their movements,. Sensors, attached with adhesive strips to doors, walls, and ceilings, can be installed in about an hour and removed in just 20 minutes, Cook says. The sensors transmit data to a small computer in the home. All information is encrypted and relayed to a server for researchers to analyze.
As a person moves around, motion sensors send the computer an “on” message. When motion stops, they send an “off” message. Because the sensors are tagged by location, the researchers can determine, for example, how many times a person enters the bathroom or walks into the kitchen, and how long they sit on the couch or stay in bed.
Magnetic sensors signal when a door is open or closed—which can help researchers determine, for example, when people leave their home or access a medicine cabinet.
Currently, the system doesn’t send doctors real-time updates of what a person is doing. But looking at the data over time can help them spot concerning signals, such as inactivity or frequent trips to the bathroom.
And over the past few years, Cook’s research team has also been studying sensor readings to determine with certainty what a person is actually doing: whether, for example, cooking, sleeping, watching television, exercising, or going out. “Once we have that understanding,” she says, “we can use what we call a Smart Home in a Box for health monitoring.”
Two things that worry seniors the most about the system, according to Cook, are whether they will lose privacy and how visible the sensors will be to visitors. To address those concerns, the researchers show them that the only data being gathered is if they moved past one point or another. And the researchers get permission for where to place each sensor.
“Most people don’t want sensors in the bathroom, for example,” Cook says. “Instead, we put them outside the bathroom door so we can infer they’re using it, but not watch them.”
The sensors cannot be hidden and are always visible, but “a few days after they’re installed, the residents forget that they’re there,” she says.
She predicts that in about five years, such smart homes will be able to recognize a diverse set of activities to assess more clearly any decline in cognitive ability or mobility. But really robust applications for the basically simple sensors—like recognizing in real time when a person is having a medical emergency—could take 15 years or so, she says.
“It’s been interesting as a machine-learning researcher to realize how complex human-behavior data is,” she says. “It’s unlike any other data I’ve studied.”