A DIFFERENTIALLY PRIVATE ENSEMBLE KALMAN FILTER FOR ROAD TRAFFIC ESTIMATION

Road traffic estimation systems can rely nowadays on an increasing number and variety of sensors and data sources to provide better coverage and accuracy, from standard static detectors to, more recently, location traces obtained possibly from individual drivers' smartphones. Motivated by privacy concerns raised by such systems, this paper discusses a methodology for estimating the macroscopic traffic state (density, velocity) along a road segment in real-time, while providing formal differential privacy guarantees to the individual drivers, a state-of-the-art notion of privacy that protects against adversaries with arbitrary side-information. The impact of the privacy constraint on estimation performance is mitigated by the use of a nonlinear model of the traffic dynamics, fused with the sensor measurements via an Ensemble Kalman Filter, a classical method for data assimilation.