Transformer-based driver identification using quantized automotive sensor data

초록

Various sensors installed in vehicles are utilized across multiple domains. Among them, real-time driver identification for resource-constrained in-vehicle platforms plays a crucial role in restricting vehicle operation to authorized users and enabling personalized control, thereby enhancing both driving safety and user convenience. This study proposes a Transformer-based driver identification model that leverages OBD-II sensor data. The model incorporates quantization-based preprocessing and majority voting to improve computational efficiency and prediction stability. Additionally, Gaussian noise-based data augmentation is adopted to evaluate the model's robustness under noisy conditions. Experiments on two public datasets show that the proposed model achieves high Accuracy and Macro F1, and remains robust under noisy conditions. Furthermore, an indirect inference latency analysis indicates that the model satisfies real-time constraints by completing predictions within the sensor input interval, validating its suitability for real-time driver identification in embedded environments. © 2026 Elsevier B.V.

키워드