TPP: Trajectory Privacy Preservation Against Tensor Voting Based Inference Attacks

The popularity of mobile devices with global positioning system (GPS) has boosted various wireless location-based services (LBSs). Certain honest-but-curious or even dishonest LBS servers may learn the users' trajectories from location trace files, and the users' privacy can be compromised. In this paper, we propose a quantitative approach to model trajectory inference attacks via tensor voting, which can be widely applied in computer vision and machine learning as a perceptual organization. To counter the tensor voting based attacks, we propose a novel trajectory privacy preservation TPP scheme, in which LBS users will intentionally generate dummy trajectories to obfuscate LBS servers. Meanwhile, the LBS users have the option to disclose their trajectories to trustworthy parties (e.g., users' parents) by sending those parties a few more encrypted locations. Considering the power constraint of hand-held mobile devices, we mathematically formulate the trajectory privacy preservation problem into a mixed integer linear programming optimization problem and propose the algorithms for optimizing solutions. Through simulations and analysis, we show that the proposed scheme can effectively preserve LBS users' trajectory privacy against tensor voting-based inference attacks with limited power consumption.