Treadmill attack on gait-based authentication systems

In this paper, we demonstrate that gait patterns of an individual captured through a smartphone accelerometer can be imitated with the support of a digital treadmill. Furthermore, we design an attack for a baseline gait based authentication system (GBAS) and rigorously test its performance over an eighteen user data-set. By employing only two imitators and using a simple digital treadmill with speed control functionality, the attack increases the average false acceptance rate (FAR) from 5.8% to 43.66% for random forest, the best performing classifier in our experiments. More specifically, the FAR of eleven out of eighteen users increased to 70% or more. Our results call for a revisit of the design of the GBAS to make them resilient to such attacks.