Fast traffic state estimation with the localized Extended Kalman Filter

Traffic state estimation is important input to traffic information and traffic management systems. A wide variety of traffic state estimation methods exist, either data-driven or model-driven. In this paper a model-driven approach is used: the LWR model solved by the Godunov scheme. The most widely applied method to combine this model with real-time data is the Extended Kalman Filter (EKF). A large disadvantage of the EKF is that it is too slow to perform in real-time on large networks. In this paper the novel Localized EKF (L-EKF) is proposed that sequentially makes many local corrections instead of one large global correction. The L-EKF does not use all information available to correct the state of the network, but in an experiment it is shown that the resulting loss of accuracy is negligible in case the radius of the local filters is taken sufficiently large. The L-EKF hence is a highly scalable solution to the state estimation problem that results in equally accurate state estimates.