A Geostatistical Approach to Traffic Flow Reconstruction from Sparse Floating-Car Data

Abstract

In this work, we consider that the underlying nature of traffic flow is a random field, and the contributions of individual vehicles, the Floating-Car Data, are single realization of those random fields. We have found that Stop-and-Go waves can be characterized by the anisotropic variogram. This function can be used to reconstruct the traffic random field with better performance than the average approach. We have tested the geostatistical interpolations (Triangular Irregular Network and Kriging) with synthetic and real-world data, scoring its performance by means of a cross-validation technique. In the future, this approach may be useful in reducing uncertainty in the estimation of travel times and even decreasing the occurrence of road capacity breakdowns.