Perceiving spatiotemporal traffic anomalies from sparse representation-modeled city dynamics

Early perception of anomaly traffic patterns, both spatially and temporally, is of importance for emergency response in the smart cities. To capture the spatiotemporal correlations among traffic flows for city dynamics modeling in correspondence with normal states, we conduct sparse representation on taxi activity over spatially partitioned cells in a city. We can perceive the deviation from the normal evolution of traffic flows and find the traffic anomalies. This method roots in the ideal of global traffic flow network detection. Therefore, it is more informative than local statistics since traffic flows evolve in a mutually interacting manner to spread out all over the city. The experimental results confirm its predictive power in detecting spatiotemporal traffic anomalies.