“Gait” refers to a pattern of locomotion achieved through the movement of limbs. Researchers have discovered that gait patterns differ among people, so much so they can be used for biometric authentication. University of Pretoria and City University of Hong Kong researchers have developed a continuous smartphone user authentication system based on gait analysis.
Currently, biometric gait authentication tools fall into three major categories:
- Machine Vision: Gait recognition based on machine vision harnesses cameras to gather data, and leaves the analysis to image processing.
- Floor Sensor: Floor sensor-based systems center on a mat that measures the speed and force of someone’s step.
- Wearable Sensor: This gait analysis method uses a wearable device fitted with sensors (gyroscopes, force sensors, accelerometers, etc.) to capture all relevant data.
However, the new system utilizes a wearable sensor-based technique, and merges that with the hardware of smartphones, specifically the accelerometer. For that reason, no additional hardware is required, and thus, no extra cost is tacked on to upgrade existing mobile devices.
“This paper presents the development of a smartphone user authentication system which takes advantage of the device’s pre-existing hardware,” the researchers wrote in their paper. “The authentication was based on a smartphone user’s gait pattern, which is a biometric feature.”
The method uses a sensor data acquisition unit, pre-processing unit, classification algorithm, and an evaluation system. The smartphone’s accelerometer constantly analyzes gait metrics, and after it is processed and analyzed, the user’s identity is confirmed. If it detects an abnormal gait change, the system sends a notification.
“If the authentication outcome is positive, the authentication process continues uninterrupted in the background,” the researchers say. “If the authentication fails, the device’s location information should be sent to a predetermined email address to notify the authorized user of the device’s whereabouts.”
In tests, the method earned a sensitivity score of 0.74 and a specificity of 0.78. Although promising, the researchers admit more development is needed until it’s ready for real-world use.
The team recently presented their paper at the 44th annual conference of the IEEE Industrial Electronics Society.