Vehicle Trajectory Clustering and Anomaly Detection at Freeway Off-Ramp Based on Driving Behavior Similarity

Freeway off-ramp safety has attracted much attention. To study motion patterns of vehicles at freeway off-ramps, a trajectory clustering method based on similarity of driving behavior features was proposed. Traffic video data was collected by an unmanned aerial vehicle and 472 vehicle driving trajectories composed of spatial position, longitudinal speed, lateral speed, speed, longitudinal acceleration, and lateral acceleration, were extracted. DTW was used to build a similarity model to measure the similarity between vehicle trajectories and a trajectory clustering method was proposed based on DBSCAN clustering algorithm. Based on trajectory clustering results, different motion patterns including regular motion patterns of driving in one lane and changing lane to adjacent lane, improper lane changing motion patterns of vehicle crossing at least one lane and illegal motion pattern of overtaking at off-ramp, are extracted. Representative trajectory of each clusters can be used to detect different trajectories and determine new abnormal trajectories.