A Systematic Analysis of External Factors Affecting Gait Identification.

Inertial sensors integrated into smartphones provide a unique opportunity for implicitly identifying users through their gait. However, researchers identified different external factors influencing the user's gait and consequently impact gait-based user identification algorithms. While these previous studies provide important insights, a holistic comparison of external factors influencing identification algorithms is still missing. In this explorative work, we conducted a focus group with participants from biometrics research to collect and classify these factors. Next, we recorded the gait of 12 participants walking regularly and being influenced by eleven different external factors (e.g., shoes and floor types) in two separate sessions. We used a Deep Learning (DL) identification algorithm for analysis and validated the analysis results using within- and between- sessions data. We propose a categorization of gait covariates based on users' control levels. Floor types have the most significant impact on recognition accuracy. Finally, between-session analysis shows less accurate yet more robust results than within-session validation and testing.