Affiliations: [a] Department of Intelligent Systems, Jožef Stefan Institute, Ljubljana, Slovenia | [b] Faculty of Electrical Engineering and Information Technology, Saints Cyril and Methodius University in Skopje, Skopje,Macedonia. E-mail: hristijan.gjoreski@ijs.si
Note: [*] The first two authors should be regarded as joint first authors.
Abstract: Person identification is a process through which a person is recognized using some information about him-/herself. Usually this is performed by asking the user to perform some action, e.g., to apply a token (card), enter a PIN code, scan a finger, or something similar. This paper describes an approach for recognizing a person entering a room using door accelerations, i.e., no additional action is required. The approach analyzes the acceleration signal in time and frequency domain. For each domain two types of methods were developed: (i) feature-based – uses features to describe the acceleration and then uses classification method to identify the person; (ii) signal-based – uses the acceleration signal as input and finds the most similar ones in order to identify the person. The four methods were evaluated on a dataset of 1005 entrances recorded by 12 people. The results show that the time-domain methods achieve significantly higher accuracy compared to the frequency-domain methods, with signal-based method achieving 86% accuracy. Additionally, the four methods were combined and all 15 combinations were examined. The best performing combined method increased the accuracy to 90%. Additional experiments with varying the number of training instances, showed that around 10 to 20 training instances are enough to achieve reliable performance. The results confirm that it is possible to identify a person entering a room using only the door acceleration and that relatively high accuracy (over 95%) is expected for a limited group of dissimilar users, e.g. a typical family.
Keywords: Person identification, door, acceleration, machine learning, dynamic time warping
