A comprehensive understanding of city structures and urban dynamics can greatly improve the efficiency and quality of urban planning and management. Traditional urban survey and statistical approaches usually incur substantial labor and time. Based on large-scale vehicle sensing data, we observe that the regular mobility patterns are correlated to urban area functionalities, while the irregular crowd movement may be caused by urban social and emergency events. In this paper, we propose a data-driven framework to sense urban structures and dynamics from large-scale vehicle mobility data in a systematic manner. First, we divide the city into fine-grained mobility grids, and propose a distance-constrained clustering algorithm (DCCA) to cluster the grids with similar mobility features into structured urban areas. Second, we detect irregular mobility traffic patterns in each area leveraging an ARIMA-based anomaly detection algorithm (ADAM), and correlate them to the urban social and emergency events. Finally, we build a visualization system to demonstrate the urban structures and crowd dynamics. We evaluate our framework using real-world datasets collected from Xiamen City, China. Results show that our approach comprehensively portraits urban structures and crowd dynamics for urban planning and city management.

/proc: data processing codes

/viz: the visualization platform