Exploring Time-dependent Traffic Congestion Patterns from Taxi Trajectory Data

Due to public travel choice, city function zoning and road network structure, urban traffic congestion tends to have strong spatiotemporal correlations. Unveiling the spatiotemporal patterns of urban traffic congestions will provide useful information for urban planning, traffic control, and location based service (LBS). This paper proposes an approach to identify traffic congestion regions and their spatiotemporal distribution from taxi trajectory data. Firstly, slow trajectory sequences are extracted from raw taxi trajectory data. Together with taxi engine states, these sequences are then transferred into congestion events that define the congestion duration and the average speed. Thereafter, highly congestion-prone areas are identified by clustering these congestion events using the DBSCAN clustering method. From the perspective of spatial homogeneity, global aggregation degrees of those identified congestion-prone areas are defined by the Ripley K function. Finally, considering congestions of nearby areas can influence each other and worsen the local traffic condition, the theory of data field is imposed to reveal the interactions between neighbouring congestion events. It also enables the visualization of the congestion intensity distribution from the trajectory potential of trajectory data field. The proposed method is validated by a case study of taxi trajectory data analysis in Wuhan City, China.