Gait-based authentication for smart locks using accelerometers in two devices

Abstract

Smart locks can be opened and closed electronically. Fingerprint or face authentication is inconvenient for smart locks because it requires the user to stop for several seconds in front of the door and remove certain accessories (e.g., gloves, sunglasses). This study proposes a user authentication method based on gait features. Conventional gait-based authentication methods have low identification accuracy. The proposed gait-based authentication method uses accelerometers in a smartphone and a wearable device (i.e., smartwatch). We extracted 31 features from the acquired acceleration data and calculated identification accuracy for various machine-learning algorithms. The highest accuracy was 95.3%, obtained using random forest. We found that the maximum interval, minimum interval, and minimum value had the highest contributions to identification accuracy, and variance, median, and standard deviation had the lowest contributions.