Visual Inertial Odometry-enhanced GNSS Spoofing Detection using Kernel Density Estimation

Reliable positioning by Global Navigation Satellite System (GNSS) is essential for safe and large-scale land transport systems operations. However, the advent of GNSS navigation has led to a heightened vulnerability of GNSS services to the spoofing attacks. Based on the involvement of the Visual-Inertial Odometry (VIO) that remains unaffected by GNSS spoofing attacks, a GNSS spoofing detection solution is proposed with the enhancement by VIO. In this solution, a Kernel Density Estimation (KDE) strategy is adopted to model the distribution of difference between VIO and non-interfered GNSS positioning results. Real-time monitoring of the deviation in the current localization difference distribution is realized by the Kullback-Leibler (KL) divergence, with which the GNSS spoofing detection logic is designed. Real-world datasets of visual/inertial/GNSS-based positioning are involved to validate the solution through GNSS spoofing injection tests. Results under two typical spoofing attack scenarios demonstrate the advantages of the proposed solution in detecting deviation caused by the spoofing attack, illustrating the great potential in the VIO-enhanced multi-sensor fusion-based autonomous positioning systems.