Generalizable Driver Identification Through Road Curvature Analysis in Internet of Vehicles

Abstract

The integration of Internet of Things (IoT) technologies into modern vehicles has facilitated the emergence of the Internet of Vehicles (IoV), revolutionizing the automotive industry by enabling advanced connectivity and data-driven functionalities. Among the many applications made possible by these advancements, accurate driver identification has become essential for enhancing vehicle security, personalizing user experiences, and supporting usage-based services. However, existing driver identification methods often struggle to maintain accuracy across various road environments, as driving behavior varies with road characteristics. This paper introduces a novel driver identification framework that dynamically adapts to varying road geometries by integrating road curvature analysis to improve both accuracy and robustness across diverse road environments. Using global positioning system (GPS) sensor data, road curvature is estimated using the Menger curvature method, and road curvature segments are classified into distinct types through k -means clustering with dynamic time warping. Separate driver identification models are then developed for each road type using machine learning algorithms, including Random Forest, XGBoost, and LightGBM, to capture the subtle differences in driving behavior with varying road types. Extensive experiments using real-world driving data demonstrate that the proposed method achieves up to 86.02% accuracy on unseen road environments and outperforms existing methods by up to 18.64%. These experimental results highlight the improved generalization capability and comprehensive validation of the proposed model, emphasizing its effectiveness for robust driver identification in realistic scenarios.