Chapter 4 - Model-based techniques for traffic congestion detection

Abstract

To ensure efficient road traffic management and improve safety, it is essential to automatically monitor road traffic congestion. This chapter is within model-based monitoring techniques. To achieve this objective, we present an enhanced observer integrating the advantages of a piecewise switched linear traffic (PWSL) model and Kalman filter (KF). Here we design the coupled PWSL-KF model based on free-flow data to mimic the uncongested traffic conditions. Subsequently, residuals generated using the PWSL-KF model are utilized by the k-nearest neighbors (kNN) to sense traffic congestions. Crucially, we are introduce two monitoring charts, kNN-based Shewhart and EWMA (exponentially weighted moving average), to set the alarm limit lines to identify traffic congestion. However, these charts are based on the assumption that the data follow a Gaussian distribution, which is often invalid in practice. To mitigate this limitation, we employ kernel density estimation to set nonparametric control limits of the proposed monitoring charts. Using traffic data from the four-lane State Route 60 in California freeways, we demonstrate the effectiveness of the developed charts in supervising traffic congestions.