Network-wide traffic flow estimation with insufficient volume detection and crowdsourcing data

With the rapid development of urbanization and modernization, it is increasingly crucial to sense network-wide traffic. Network-wide traffic volume information is of great benefit for traffic planning, government management and vehicle emissions control. However, it is difficult to install detectors on every intersection due to the expensive deployment and maintenance costs, and the insufficient sensor coverage across the network limits the direct availability of network-wide traffic flow information. Whereas, crowdsourcing floating car data with a high coverage rate are currently available, which creates an opportunity to address this problem. In this paper, we propose a novel methodology to estimate network-wide traffic flow, which incorporates flow records and crowdsourcing floating car data into a geometric matrix completion model. Furthermore, a spatial smoothing index based on the divergence is developed to measure the difficulty of volume estimation for each road segment. We conduct extensive experiments on both real-world and synthetic datasets. The results demonstrate that our approach consistently outperforms other benchmark models and that the proposed index is highly correlated to estimation accuracy.