3 minute read
An Excerpt from “Gesturing with Smart Wearables: An Alternate Way to User Authentication”
Khandaker Abir Rahman
Associate Professor of Computer Science & Information Systems
Avishek Mukherjee
Assistant Professor of Computer Science & Information Systems
Kristina Mullen
SVSU Computer Science Major and Cybersecurity Minor (Class of 2022)
Khandaker Abir Rahman received his Ph.D. and M.S. in computer science and his M.S. in mathematics from Louisiana Tech University, and his B.S. and M.S. incomputer scienceandengineeringfrom theUniversityofDhaka,Bangladesh. His research interests include behavioral biometrics, cybersecurity, machine learning, and artificial intelligence. He has co-authored and/or presented 25 research articles in refereed journals and/or at international conferences, and produced five book chapters. Dr. Rahman has also been working as a reviewer for several distinguished journals. He is SVSU’s representative to the Michigan Space Grant Consortium. In 2021, he was awarded senior member status by the Association of Computing Machinery (ACM) and Institute of Electrical and Electronics Engineers (IEEE).
Avishek Mukherjee received his Ph.D. and M.S. in computer science from Florida State University. His research interests include wireless networks and systems with an emphasis on physical layer algorithms. He has been awarded several internal research grants at SVSU as well as a research seed grant from the Michigan Space Grant Consortium in 2020. In addition to his research, Dr. Mukherjee also serves as the faculty advisor for SVSU’s chapter of the Association of Computing Machinery.
Kristina Vargo née Mullen received her B.S. from SVSU in 2022. She was awarded the Michigan Space Grant Consortium’s undergraduate research fellowship for the 2021–2022 school year for proposed research work in the field of cybersecurity.
“Gesturing with Smart Wearables: An Alternate Way to User Authentication” first appeared in 2022 in ITNG 2022 19th International Conference on Information Technology-New Generations, Advances in Intelligent Systems and Computing 1421, edited by Shahram Latifi. The full article can be found at https://doi.org/10.1007/978-3-030-97652-1_17.
Abstract
A method of alternate user authentication that relies on sensory data from a smartwatch has been explored in this paper. This attempt to beef up the authentication security was made by taking the user-defined hand gesture into account while wearing a smartwatch. Eventually, the preset hand gesture would work in a similar way to the password-based authentication scheme. In our experiment, we recorded the 3D coordinate values measured by the accelerometer and gyroscope over a set of gestures. We experimented with 50 gesture samples comprising of five different gesture patterns and ten repeated samples for each pattern. We developed an Android WearOS smartwatch app for sensor data collection, implemented our method of sensor data processing, and performed a series of experiments to demonstrate the potential of this method to achieve high accuracy
3 Data Processing and Experimental Setups
3.1 Pre-Processing of Data
As mentioned in the previous section, the movement data was collected over several days to add some randomness to the movement. A by-product of this collection process was that there was a timing offset for most of the signals collected. To compensate for this, a stationary period of six seconds was recorded at the start and end of the recorded movement. When comparing two signals, the stationary period above was used to synchronize the signals before computing the squared error between them. The alignment was done using a threshold heuristic based on standard deviation from the first three seconds of recorded sensor data on both sensors. The starting position of the movement was identified by setting the threshold at least ten times the threshold found in the first three seconds (i.e., stationary period). A similar strategy was implemented to identify the end of the recorded movement by looking at the last three seconds of sensor data. Finally, the process was repeated across all three axes data reported by each sensor, and the final starting and ending timestamps were determined by looking at the mean values reported by every axis. In addition, the signal amplitude was normalized based on the largest amplitude measured across all axes. An example of the alignment result can be seen in Fig. 1, where the amplitude of the signal measured on all three axes on the accelerometer is shown. The vertical barriers indicate the subset of the signal data that is clipped and considered during the pre-processing stage. It can be seen that the alignment method described above works very well to identify the start and end times of the movement.
3.2 Error Computation
To compare the reference and the authentication/test samples, the difference between the two sensor signals was computed. First, the length of signals was considered, and the longer of the two signalswastruncatedtomatchthelength oftheshorter signal. Thenthepair-wise squared error between each corresponding amplitude was computed for all axes, and finally, the total difference reported by the accelerometer sensor for comparison was computed. Likewise, an identical process can be used to represent the total difference based on the gyroscope data.
As an example, Fig 2 and Fig 3 show the computational error when using the accelerometer data across two sets of signal pairs. Fig 2 shows the aligned signals on each of the three axes when comparing a pair of signals that belong to the same pattern. Naturally, after correct alignment, the signal data follow very closely to one another, with a very low error value. On the other hand, Fig 3 shows the aligned signals when looking at signal data from different patterns.
Reprinted by permission of the authors and Springer International Publishing AG, part of Springer Nature, 19th International Conference on Information Technology-New Generations ITNG, 2022.