Biometric Gait Authentication Using Accelerometer Sensor

Abstract — This paper presents a biometric user authentication based on a person’s gait. Unlike most previous gait recognition approaches, which are based on machine vision techniques, in our approach gait patterns are extracted from a physical device attached to the lower leg. From the output of the device accelerations in three directions: vertical, forward-backward, and sideways motion of the lower leg are obtained. A combination of these accelerations is used for authentication. Applying two different methods, histogram similarity and cycle length, equal error rates (EER) of 5% and 9 % were achieved, respectively. Index Terms — security, biometric, gait recognition, sensorbased gait, unobtrusive authentication, leg acceleration