Proposal and Evaluation of a Personal Authentication Method Using Insole-Type Gait Sensors

This study proposes a novel method for 1:1 personal authentication using asymmetry of gait data obtained from insole-type gait sensors. This approach aims to facilitate authentication without the need for conscious actions, thus enabling seamless, unconscious user identification. The challenge in personal authentication using insole-type gait sensors arises from the variability in walking speed, which is rarely constant in daily life. Consequently, it is crucial to ensure that the gait data, central to the authentication process, remains consistent despite varying walking speeds. To address this issue, gait data was visualized and analyzed at different speeds to identify elements, the asymmetry of gait data, that are robust to changes in walking speed. Based on these findings, machine learning models were trained on the asymmetry of gait data to authenticate individuals. Furthermore, to improve both authentication accuracy and robustness, majority voting based on the authentication results of multiple steps is utilized. The effectiveness of the proposed method was assessed by evaluating authentication accuracy using metrics such as False Acceptance Rate (FAR), False Rejection Rate (FRR), and Equal Error Rate (EER). The evaluation results demonstrate that our method can identify individuals with a reasonable degree of accuracy, highlighting its potential for facilitating unconscious authentication.