Estimating the posterior predictive distribution of the traffic density in multi-lane highways using spacing measurements

Traffic state estimation (TSE) is an important task for traffic management, however it can be challenging due to the sparse deployment of traffic sensors in the network. The advancement of new vehicle technologies provides new potentials for TSE using extended floating car data (xFCD). In this work we propose a probabilistic approach that makes use of xFCD, which consists of information such as the position and spacing of individual vehicles, collected by Connected and Automated Vehicles (CAVs) deployed in the traffic network under study. The proposed methodology takes into account spacing information from each CAV and utilises the Bayesian paradigm along with a maximum-a-posteriori (MAP) plug-in estimate to derive the traffic density of a highway. The proposed methodology, referred to as MAP-TSE, is evaluated using a real-life dataset extracted from videos recorded by Unmanned Aerial Vehicles (UAVs). Results presented in this work illustrate efficient estimation of the traffic density for low penetration rates and different time-window sizes, yielding lower error compared to the literature approach.