Researchers Use AI to Identify People via Seismic Signals Produced from Footsteps

Biometric sensors have been used to identify people through their physiological characteristics for some time now. Authorities use everything from iris and fingerprint scans to voice imprints to identify individuals, and thanks to some clever Indian Institute of Technology researchers, they can add seismic data generated by walking to that toolbox.

The researchers describe their new platform in a paper titled “Person Identification using Seismic Signals Generated from Footfalls,” which employs the AI-based fog computing architecture that uses edge devices for computation and communication, and stores that data in the cloud. The team notes that most biometric systems require active cooperation from the individual for identification, while all that’s needed for their system is for the individual to walk through a specific geometric location.

The researchers designed their seismic footfall system using two platforms — a sensor system and a fog computing unit. The sensor system features a geophone (seismic sensor) to garner footfall data, a Raspberry Pi Zero and sound card to process it, and an Xbee 868 LP/antenna to transmit that data to the fog computing unit, which utilizes a Raspberry Pi 3, and another Xbee to send the info into the cloud.

To train their platform to identify different steps from local individuals, the researchers used eight barefoot participants and collected the time and frequency of their footfalls, as well as their cadence and length. Over a one-month period, they used their system to collect 46,000 footfall events from the participants and found that it took only eight minutes (or roughly 875 footsteps) for the system to identify specific individuals at an accuracy of 92.29%.

While that’s certainly an impressive feat for their seismic footfall platform, it does have one notable drawback, in that it can only identify one person at a time, otherwise the system gets confused if there are two or more. Of course, they plan to rectify that issue in a future revision but state that it can already be used for applications such as registering school attendance, detecting home intruders, or even to automate home appliances.