A combination of microphones and artificial-intelligence algorithms can identify whether a person in a room is sitting, standing, walking or falling.
The use of deep-learning algorithms for identifying human activity from video feeds has promising applications, such as alerting caregivers of a medical emergency. But continuous video surveillance raises the possibility of leaks, hacks and loss of privacy.
As an alternative to video, a team led by Xinhua Guo at the Wuhan University of Technology in China and Kentaro Nakamura of the Tokyo Institute of Technology turned to high-frequency sound waves. The researchers designed an acoustical array with four speakers that emit a signal of 40 kHz — above the range of human hearing — into a room. Surrounding the speakers are 256 small microphones that pick up the high-pitched tones reflected back by the environment.
The team used the array to track volunteers as they sat, stood, walked or fell. After deep-learning algorithms were trained on the reflected high-frequency sounds, the programs could identify an individual’s activity with up to 97.5% accuracy.
Monitoring devices like these, which rely on high-frequency sounds, could assuage privacy concerns, the authors write.