Zero-to-Stable Driver Identification: A Non-Intrusive and Scalable Driver Identification Scheme

Driver identification faces various challenges in real-time applications. These challenges include high dimensional input data, moderate accuracy, scalability issues and need for custom-built in-vehicle sensors. The conventional biometric solutions are either less accurate, unscalable, or costly for practical application. In those solutions, high accuracy comes at the cost of the preservation of privacy. Similarly, scalability and cost-effectiveness are inversely proportional to one another. This paper proposes a driver identification scheme to pare these challenges. The scheme uses data from the global positioning system (GPS) to learn an individual's driving pattern, which is commonly deployed in in-car navigation systems and can also be found in general-purpose hand-held devices in the prevailing market. The proposed scheme is innovative in terms of providing a single solution for the existing challenges. It is more practical and scalable with large numbers of the drivers also being cost-effective and accurate, which makes it applicable in real-time applications with the least overhead costs for different resources. The consequent analysis and empirical results show that the scheme could identify drivers with significant accuracy given only GPS data. To assess the scheme in-depth, we perform experiments both on the collected data and the real-world open datasets. The average accuracy approximates above 96% for up to 25 drivers.