Return to Session B4b

Session B4b: Sensor Fusion

GNSS Spoofing Detection Using Visual Inertial Odometry
Xiao Zhou, Hong Li, Jian Wen, Yimin Wei, Dept. of Electronic Engineering, Tsinghua University, China; Chun Yang, China Academy of Engineering Physics, China; Mingquan Lu, Dept. of Electronic Engineering, Tsinghua University, China
Location: Beacon B

Peer Reviewed

Peer Reviewed

GNSS Spoofing Detection Using Visual Inertial Odometry
Introduction
Global navigation satellite system (GNSS) spoofing is becoming an emerging threat to GNSS security for it can induce fake positions at the victim receiver. Numerous anti-spoofing techniques have been proposed to shelter the receivers from spoofing attacks. However, with the development of vision/GNSS integration platforms in recent years, the safety of the GNSS signals lies as a crucial issue. This paper proposes a spoofing detection method based on the open-source vision/GNSS integration platform VINS (Visual-Inertial Navigation System)-Fusion. The first part of the proposed method is based on the consistency check of GNSS measurements and multi-sensor state estimation, while the second part of the proposed method is based on the transformation matrix of the local frame and the global frame. Both methods avoid a fixed inaccurate coordinate transformation, making the proposed method more reliable than the vision-based spoofing detection methods being compared. In this paper, we first test the performance of VINS-Fusion. Then, simulations results show the effectiveness of our spoofing detection method in spoofing scenarios with an acceptable false alarm probability and missing alarm probability.
Performance of VINS-Fusion
In this paper, we evaluate the performance of VINS-Fusion to verify the necessity to implement spoofing detection in a vision-based state estimation system and explore whether VINS-Fusion can be used to detect spoofing.
We evaluate the accuracy of the vision-based estimation in VINS-Fusion by comparing the trajectory generated by VIO and GPS. The comparison results show that the VIO and the GPS trajectories are highly coincidental in the x and y-axis. However, the VIO produces a more significant error in the z-axis, which means the vertical direction. And the error accumulates over time. We verified it on other datasets, and VINS-Fusion did show similar performance in the z-axis. Meanwhile, we compare the GPS trajectory and the fused trajectory to evaluate the accuracy of the coordinate transformation in VINS-Fusion. We noticed that the fused trajectory has a period jitter in the vertical direction, and the period equals the length of the optimization sliding window. According to our analysis, this may be due to the vertical error in VIO. The comparison results show that the transformation in the x-y plane is more accurate, which means the Euler angle of the z-axis we get from the transformation matrix is more accurate than those of the y and x-axis.
Moreover, we evaluate the performance of VINS-Fusion when facing GNSS spoofing by slightly modifying the GPS positioning of the dataset. The result of the fused trajectory shows that the accuracy of the fused trajectory largely depends on the accuracy of the GPS positioning, which means VINS-Fusion is vulnerable under GNSS spoofing. The test result emphasizes the vulnerability of GNSS-related navigation systems and the necessity to implement GNSS spoofing detection.
GNSS Spoofing Detection Based on The Consistency Check of Different Sensors
The first part of our spoofing detection method is based on the consistency check of GNSS measurements and multi-sensor state estimation provided by VINS-Fusion. The test is performed on each new GNSS measurement. We set a fixed observation window and compare the trajectories provided by GNSS and VIO in the window when receiving each new GNSS measurement. For the spoofing scenarios where the GPS trajectory of the victim receiver is slightly modified, the simulation results have indicated that the proposed method is realizable for this spoofing scenario.
GNSS Spoofing Detection Based on The Transformation Matrix
The GNSS spoofing detection method based on the consistency check of different sensors works in most spoofing scenarios. However, it is not applicable when the spoofer knows the mean velocity of the victim receiver and considers this when generating the fake GNSS signals. We proposed a detection method based on the transformation matrix to solve this issue.
According to our analysis of the transformation matrix of VINS-Fusion, the transformation in the x-y plane is more accurate, which means the change of the Euler angle of the z-axis has the potentiality to indicate the occurrence of a spoofing attack. We defined another detection statistic which is associated with the rotation around the z-axis used to parameterize the transformation matrix. For the spoofing scenarios where the GPS trajectory of the victim receiver is slightly rotated, the proposed method can detect the spoofing when the rotation angle is larger than 3°.
CONCLUSIONS
In this paper, we proposed two spoofing detection methods. For the spoofing scenarios where the mean velocity of spoofing signals is different from the real velocity of the vehicle, the first method takes advantage of the inconsistency of the received GNSS signals and the multi-sensor state estimation. It can identify the occurrence of spoofing attacks. For the spoofing scenarios where the spoofer knows the mean velocity of the vehicle and considers this in the generation of fake GNSS signals, the second method takes advantage of the optimization of VINS-Fusion and can detect the spoofing when the rotation angle of the trajectory is larger than 3°.To sum up, these two methods use the inconsistency of the trajectory generated by GNSS and VIO. For the second method, different from other methods which use fixed transformation matrix to compare the trajectories, we get a changing transformation matrix in every optimization of VINS-Fusion and use the change of the matrix to evaluate the spoofing. Simulation results show the effectiveness of the proposed method in different spoofing scenarios.



Return to Session B4b