Detection of Anomalous Traffic Patterns and Insight Analysis from Bus Trajectory Data

Detection of anomalous patterns from traffic data is closely related to analysis of traffic accidents, fault detection, flow management, and new infrastructure planning. Existing methods on traffic anomaly detection are modelled on taxi trajectory data and have shortcoming that the data may lose much information about actual road traffic situation, as taxi drivers can select optimal route for themselves to avoid traffic anomalies. We employ bus trajectory data as it reflects real traffic conditions on the road to detect city-wide anomalous traffic patterns and to provide broader range of insights into these anomalies. Taking these considerations, we first propose a feature visualization method by mapping extracted 3-dimensional hidden features to red-green-blue (RGB) color space with a deep sparse autoencoder (DSAE). A color trajectory (CT) is produced by encoding a trajectory with RGB colors. Then, a novel algorithm is devised to detect spatio-temporal outliers with spatial and temporal properties extracted from the CT. We also integrate the CT with the geographic information system (GIS) map to obtain insights for understanding the traffic anomaly locations, and more importantly the road influence affected by the corresponding anomalies. Our proposed method was tested on three real-world bus trajectory data sets to demonstrate the excellent performance of high detection rates and low false alarm rates.