Classification of acceleration data for biometric gait recognition on mobile devices

Abstract

Ubiquitous mobile devices like smartphones and tablets are often not secured against unauthorized access as the users tend to not use passwords because of convenience reasons. Therefore, this study proposes an alternative user authentication method for mobile devices based on gait biometrics. The gait characteristics are captured using the built-in accelerometer of a smartphone. Various features are extracted from the measured accelerations and utilized to train a support vector machine (SVM). Among the extracted features are the Meland Bark-frequency cepstral coefficients (MFCC, BFCC) which are commonly used in speech and speaker recognition and have not been used for gait recognition previously. The proposed approach showed competitive recognition performance, yielding 5.9% FMR at 6.3% FNMR in a mixedday scenario.