How Machine Learning is Making Independent Living Safer for the Elderly

July 23, 2018

Seniors often appreciate the ability to stay in their homes and live independently. This independence isn’t without its challenges; the biggest being the threat of falling or having a health episode at home without anyone knowing. Fortunately, this is an area in which artificial intelligence and machine learning stand to make a difference, and soon.

Looking back, early fall alert buttons were comically large and cumbersome to wear, though the concept was good. Today’s versions are an improvement – they’re smaller and easier to carry, making them more likely to be used. Still, some of the most exciting potential is in wearable and non-wearable sensors that are fully automatic and don’t require the carrier to take any action.

For example, a study from the University of Warwick focused on the ideal placement of inertial sensors on the body to identify fallers using AI. In their walking test, researchers found that the most effective way to assess fall risk is to use the velocity measurement picked up by a sensor placed on the shins. Meanwhile, for detecting falls for those standing still or trying to stand up (or sit down), linear acceleration data from a sensor on the lower back is the most accurate.

Insights like these will help shape the wearable sensors of the future and make them more accurate.

Another route is the non-wearable system, which uses sensors in a senior’s environment to detect their status at any given time. Though some seniors may view the monitoring as creepy, others like it because it’s less obtrusive. Non-wearable systems can be based on one of four inputs: ambient noise (vibrations from falls), WiFi (signal strength), vision (images from a camera) and radio frequency identification (RFID – using the Doppler frequency value).

The first three inputs are either too complex or not accurate enough, according to researchers from Nanjing University. Thus, they set out to refine a method for using a passive RFID tag hanging around the subject’s neck to spot falls. In their study, they found that, focusing on Doppler frequency, “the value of a sudden fall was significantly higher than that of the other movements,” enabling them to distinguish the falling action based on the change in speed.

By establishing which data is meaningful and consistent (in this case, readings from the RFID tag), researchers can train AI systems to have a much better sense of when an emergency is taking place.

To learn more about how artificial intelligence is improving care for people in their golden years, read our 2018 study on Generation AI.