First real-world study shows the potential of gait authentication to enhance smartphone security

UNIVERSITY OF PLYMOUTH

Real-world tests have shown that gait authentication could be a viable means of protecting smartphones and other mobiles devices from cyber crime, according to new research.

A study led by the University of Plymouth asked smartphone users to go about their daily activities while motion sensors within their mobile devices captured data about their stride patterns.

The results showed the system was on average around 85% accurate in recognising an individual’s gait, with that figure rising to almost 90% when they were walking normally and fast walking.

There are currently more than 6.3billion smartphone users around the world, using their devices to provide a wide range of services and to store sensitive and confidential information.

While authentication mechanisms – such as passwords, PINs, and biometrics – exist, studies have shown the level of security and usability of such approaches varies considerably.

Writing in Computers & Security, the researchers say the study illustrates that – within an appropriate framework – gait recognition could be a viable technique for protecting individuals and their data from potential crime.

Academics from Plymouth’s Centre for Cyber Security, Communications, and Network Research have been focused upon developing a range of innovative authentication mechanisms to provide more secure and usable solutions.

This study builds upon that prior work through the evaluation of a multi-algorithmic gait recognition system and are the first to apply this using real-world data.

For the research, 44 participants aged between 18 and 56 were each asked to carry a globally available smartphone device for seven to 10 days.

They were asked to place the smartphone in a belt pouch to record the sensor data captured by the device’s gyroscope and accelerometer during the course of different physical activities.

Each participant generated an average of 4,000 sample activities during the course of the test, with these split into records showing normal and fast walking in addition to climbing and descending stairs.

This showed a potential error rate of 11.38% and 11.32% for normal and fast walking respectively, with the figure rising to 24.52% and 27.33% when participants were going down and upstairs respectively.

The researchers say this emphasises the need to further advance the ability to automatically differentiate a wider set of walking activities so that a multi-algorithmic approach to identification can target specific walking characteristics.