Walk the walk: attacking gait biometrics by imitation

摘要: Since advances in gait biometrics are rather new, the current volume of security testing on this feature is limited. We present a study mimicking, or imitation, human gait. Mimicking very intuitive way attacking biometric system based gait, and still topic almost nonexistent open literature. The bottom line question our research weather it possible to learn walk like someone else. If turned out be easy, would have severe effect potential as an authentication mechanism future. We developed software tool that uses wearable sensors collect analyze acceleration data. further experiment, involving extensive training test subjects, using various sources feedback video statistical analysis. attack scores analyzed by regression, goal determine whether not participants increasing their mimicking skills, simply: if they learning. The experiment involved 50 enrolled into system. error rates compete with state art technology, EER 6.2%. part revealed difficult task, physiological characteristics work against us when we try change something fundamental walk. showed few indications learning, results most attackers even worsened over time, showing did nothing help them succeed. The identified natural boundary impostors' performance, point resistance so significant was given name; plateau. location value plateau predetermines outcome attack; for success has lie below acceptance threshold corresponding Equal Error Rate (EER).