LIFT the AV: Location InFerence aTtack on Autonomous Vehicle Camera Data

Connected and autonomous vehicles (CAVs) are one of the main representatives of cyber-physical systems (CPS), where the digital data generated in several forms, such as geolocation, distance, and camera data, are used for the physical functionality of the vehicles. The utility of these data, especially camera data for computer vision projects, has contributed to the advancement of high-performance cyber-physical applications. However, location inference attacks, which involve extracting knowledge from camera data to track and estimate user locations are potential privacy threats to AV-generated camera data. In this paper, we propose LIFT (Location InFerence aTtack), a robust geo-localisation technique to exploit subjects' location privacy in a distorted GAN-based (Generative Adversarial Network) camera dataset. LIFT improves image matching of distorted query images by formulating a distinctive image nearest neighbour selection with the scale-invariant feature technique (SIFT) for feature detection and optimised pairwise clustering technique. We evaluate the performance of LIFT on the 200k Google street-view data as the reference data and 500 distorted image data (using the data generated from the Auto-Driving GAN technique) as test query data. We show that the localisation accuracy of LIFT outperforms the benchmark techniques by 20%