GPS Trajectory Biometrics: From Where You Were to How You Move
In this paper, we study how well GPS data can be used for biometric identification. Previous work has considered only the location and the entire route trajectory pattern. These can reveal the user identity when he repeats his every day moving patterns bu
- PDF / 422,912 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 39 Downloads / 223 Views
Abstract. In this paper, we study how well GPS data can be used for biometric identification. Previous work has considered only the location and the entire route trajectory pattern. These can reveal the user identity when he repeats his every day moving patterns but not when traveling to new location where no route history is recorded for him. Instead of the absolute location, we model location-independent micro movements measured by speed and direction changes. The resulting short-term trajectory dynamics are modelled by Gaussian mixture model - universal background model (GMM-UBM) classifier from speed and direction change features. The results show that we can indentify users from OpenstreetMap data with an equal error rate (EER) of 19.6 %. Although this is too modest result for user authentication, it indicates that GPS traces do contain identifying cues, which could potentially be used in forensic applications.
1
Introduction
Thanks to smart devices combined with an increasing number of social media applications, collecting and sharing of personal data has never been easier as it is today. Besides photo and video uploads, smart-phones provide direct or processed information of the user’s location or behavior via global positioning system (GPS), accelerometer or other sensor data. As an example, a sportswoman might upload her running route coordinates along with physical performance data. GPS coordinate data contains a rich source of information about the user’s whereabouts and behavior. This information can be used to provide useful services such as recommending potential friends based on user’s trajectories [17]. On the other hand, it also raises a question of privacy [4]. Location-related or spatial cues include the most commonly used locations (such as user’s home) or routes (such as daily route from home to work). Once combined with temporal (time-stamp) information, one is able to, for example, infer the future movements of a user [14] and the most likely times she will be absent from her home [3]. Speed estimates can be used for inferring the most likely means of transport (walking, bicycling, driving) [15] or whether the user respects speed limits. c Springer International Publishing AG 2016 A. Robles-Kelly et al. (Eds.): S+SSPR 2016, LNCS 10029, pp. 450–460, 2016. DOI: 10.1007/978-3-319-49055-7 40
GPS Trajectory Biometrics: From Where You Were to How You Move
451
Even if a GPS trajectory data would be anonymized by obscuring the obvious identifying information, such as name and home address, the user might still be re-identifiable by linking an anonymized GPS coordinate data with nonanonymized data in the user’s public profile in a social media application. Such information could be very sensitive; examples might be visit to an abortion clinic, church or premises of a political party [10]. In this study we focus on user identification based on GPS trajectory data. Differently from prior work that use user’s location history for identification (where you were) [10,13], we approach the problem as a biom
Data Loading...
