Privacy Protection of Vehicles Location in Spatial Crowdsourcing under Realistic Adversarial Models

Anna Squicciarini

Sponsoring Agency
National Science Foundation


With ubiquitous wireless connectivity and continued advances in positioning technologies in mobile devices, vehicle-based spatial crowdsourcing (VSC) is emerging as a new model of crowdsourcing to enable requesters to outsource their tasks to a group of vehicles (workers), who are required to physically move to tasks' locations to perform the tasks. To ensure tasks to be assigned to the nearby vehicles, vehicles are expected to disclose their location information to VSC servers, which may lead to serious privacy concerns.

A widely used location privacy protection paradigm is based on location obfuscation, which allows mobile users to upload noisy locations to servers. However, most existing location obfuscation methods assume users' mobility on a 2-dimensional (2D) plane, wherein vehicles (or workers, in case of VSC) can move in arbitrary directions at random speed without any restriction, and ignore vehicles’ mobility patterns. The additional auxiliary information obtained by considering road networks and vehicles patterns increases the risk of location exposure. Further, location obfuscation based on 2D planes is likely to generate low quality of service (QoS).

Accordingly, this research project seeks to address these issues by developing a new location privacy framework for workers in VSC. The project will start with developing new adversarial models to capture the network-constrained mobility features of multiple vehicles operating over roads. Vehicles' mobility will be described by a Bayesian network, i.e., the exact and reported locations from vehicles are considered as hidden and observable states, respectively, and the spatial correlation between hidden states will be learned from the road network environment and traffic flow information. As a countermeasure of the adversarial models, the project will develop a new location obfuscation paradigm that can effectively protect vehicles' location privacy even assuming that adversaries use vehicles' mobility features for inference attacks. Considering that the impact of location obfuscation on both privacy level and QoS may vary significantly over different road segments, the proposed location obfuscation approaches will be adaptive to various local road network conditions.

Research Area
Artificial Intelligence and Big Data
Privacy and Security