The sound wave research could have major implications for medical alarms, especially for senior citizens living alone
Spotted: Chinese researchers have discovered a way to use sound and deep learning to detect and correctly interpret human movement. The work could result in health detectors that are less intrusive than systems that rely on image sensors, they say.
The research team, based at Wuhan University of Technology, says its work allows for more accurate human activity recognition (HAR). HAR is a type of deep learning that predicts and analyses human behaviour. It is used in video surveillance, healthcare, smart homes, and sports.
In a recent paper, the research team noted that audio or visual sensors are commonly used in HAR. Visual sensors are accurate, but raise privacy concerns. In addition, visual sensors struggle to deal with smoke, low light and other challenges. But this method uses sound waves for more accurate results. Poor visibility does not affect sound waves.
Audio, or sound waves, have been less accurate in the past due to the limited number of sensors used. The Wuhan University team used 256 receivers and four ultrasonic transmitters to collect data. It also used three-dimensional (3D) acoustic data. The researchers were able to gather data connected to four common human activities: sitting, standing, walking and falling.
They used an AI algorithm to identify each type of motion based on the data. Their method resulted in up to 99% correct recognition rate with sound waves, the team says.