Lightweight Continuous Authentication via IMU Fingerprinting for V2X

Inertial measurement unit (IMU) fingerprinting is a promising physical authentication technique based on hardware imperfections produced during sensor manufacturing. This paper presents a two-stage feature extraction process that combines feature selection and mapping; the proposed approach is tailored for the lightweight vehicle-to-everything (V2X) application scenario. Specifically, the selected features are transformed into images via Gramian angular difference field (GADF), Gramian angular summation field (GASF), and Markov transition field (MTF) mappings, as well as feature extraction implemented via a convolutional neural network (CNN). Owing to the advances provided by the proposed scheme, a lightweight feature extraction system achieves satisfactory accuracy levels above 99.10% with fewer sample data and a short training time. The effectiveness and robustness of the developed approach were validated under various driving conditions via 20 IMU sensors, Arduino, and a Raspberry Pi across 20 vehicles. Additionally, tests conducted across different deep learning models demonstrated the generalizability of the proposed preprocessing and mapping methods.