Understanding urban dynamics based on pervasive sensing: An experimental study on traffic density and air pollution

Modern urban areas are influenced by numerous inter-related human and natural factors and present extremely complicated non-linear and dynamic properties. As a bridge between the physical world and the cyber space, pervasive sensing technologies make it possible to have a deep understanding of urban environments. There are two inter-related technical planes to achieve it, namely, sensing and computing. In this paper, we introduce our work progress on the urban dynamics study for each technical plane, respectively. First, we design and implement a prototype urban sensing system based on automobiles as mobile agents, performing information-rich data collection. Then, we propose a Spatial–Temporal-Entangled Analysis (STEA) algorithm based on semi-supervised manifold learning and the regularized optimization to extract semantic information from incomplete sensing data, aiming for better understanding of the spatial–temporal correlation of a complex physical process and the correlation between human activities and environmental changes via incomplete sensing data. Finally, we evaluate STEA in a real urban sensing application. Specifically, the proposed prototype is used to collect traffic and air pollution data in Beijing, and such real-world datasets are used to evaluate the effectiveness of STEA. The results obtained are very promising and show an implicit correlation between the traffic density and the air pollution, demonstrating the potential of this technique in environmental studies for urban areas.