In vehicle-based spatial crowdsourcing (VSC), requesters can outsource their tasks to a group of vehicles, which are required to physically move to tasks' locations to perform services or tasks. To promote a cost-effective task distribution, vehicles need to disclose their location information to VSC servers. Location sharing however raises serious privacy concerns related not only to whereabouts of the vehicles but also to sensitive information such as drivers’ home/working address, sexual preferences, financial status, etc. Current privacy protection mechanisms for location-services include location obfuscation methods according to mobility patterns projected on a 2-dimensional plane, wherein users can move in arbitrary directions without any restriction. Obfuscation algorithms based on a 2-dimensional plane are unable to provide strong privacy guarantees of vehicles whose mobility is restricted by road networks, since road networks and traffic patterns facilitate vehicle tracking and trajectory estimation. This research project aims to develop new location privacy protection techniques by considering vehicles’ realistic mobility features, and consequently lead to a more secure and trustworthy computing environment in VSC. This project paves the way for a more realistic body of work on location privacy, particularly regarding location-based services (LBSs). As privacy concerns are still among the main obstacles for mobile users to participate in many advanced LBSs, this project is poised to contribute to the wider adoption of LBSs for many applications (e.g. location-based recommendation systems). In addition, the project provides a set of diverse and interesting topics for undergraduate and graduate students and outreach activities for the community.
The project consists of three tasks. First, the project starts with developing new adversarial models to capture the network-constrained mobility features of multiple vehicles operating over roads. Vehicles’ mobility is described by a Bayesian network, i.e., the exact and the reported locations of vehicles are considered as hidden and observable states, respectively, and the spatial correlation between hidden states can be learned from the road network environment and traffic flow information. Second, as a countermeasure for the adversarial models, the project develops a new location obfuscation paradigm that can effectively protect vehicles' location privacy without compromising quality-of-service (QoS), even assuming that adversaries can leverage vehicles’ mobility features for inference attacks. Since the impact of location obfuscation on both privacy level and QoS vary significantly over different road segments, the new location obfuscation methods are designed to be adaptive to various local road network conditions. Finally, considering the scalability and the dynamics of VSC, the project applies distributed and parallel computing techniques (e.g., optimization decomposition) to guarantee the obfuscation algorithms to be implemented in a time-efficient manner.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
