Estimation of macroscopic traffic variables in cities with big but sparse multi-sensor data

Abstract

As urban centers become persistently denser and wider, research on realistic macroscopic models has steadily gained momentum over the last decades. Many efforts have been carried out on the modeling of traffic in urban environments, but also on the suitability of such models for real time congestion management and their potential policy implications. Surprisingly however, there are very few works in the literature that aim at estimating the macroscopic variables that are used as inputs for all other applications (e.g. real-time control, dynamic routing). Keyvan-Ekbatani et al. (2013), Ortigosa et al. (2013) and Leclercq et al. (2014) have studied how the number of detectors in a city or their location within the links can influence the quality of estimates for different applications. While this is certainly useful in the case where new detectors can be added in a network, in some other cases cities already have their roads instrumented with detectors and might not be willing to invest in a significant expansion of their network coverage. Fortunately, the emergence of smartphones and embedded systems nowadays provide rich amounts of traffic data that can be used as an additional source of measurements. On one hand, probe vehicles (also called Lagrangian sensors) travel over the entire network but have a highly time-dependent penetration rate; this creates a significant uncertainty for this type of measurements. On the other hand, inductive loop detectors (Eulerian sensors) are in fixed locations and therefore only provide a partial image of the network. Thus, the challenge of the current work is to combine both sources of data in order to provide low-uncertainty and unbiased estimates of macroscopic traffic variables (i.e. vehicle accumulation and flow) in real-time. There are a few works that deal with traffic estimation in highways (Patire et al., 2015), or with travel time estimation in urban networks (Hofleitner et al., 2012) but to the best of our knowledge, the combination of Eulerian and Lagrangian sensors to estimate accumulation and flow in urban networks remains unstudied. The methodology presented here applies a classical Bayesian framework for real-time data fusion of urban multi-sensor data. Finally, the estimation scheme is tested in a micro-simulation environment for which the ground truth is known.