Abstract

The high-resolution high-accuracy time-series trajectory data preserves rich motion status and characteristic information of vehicles in a microscopic manner. In order to analyze novel driving events or broadcast potentially hazardous situations to nearby road users, the initial step is how to identify them. Traditional approaches of manually checking trajectory datasets are extremely time-consuming and error-prone. Therefore, how to identify very few anomalous or atypical vehicle trajectories efficiently and effectively from a huge number of regular or typical trajectories is a problem that needs to be solved. This paper introduces an algorithm to automatically identify anomalous trajectories from recorded trajectory datasets in the order of trajectory clustering, template trajectory extraction, trajectory similarity comparison, and anomaly detection. A Dynamic Time Warping (DTW)-based hierarchical clustering model is leveraged to extract critical template trajectories from historical trajectories. By comparing the distance similarity with the templates, trajectories can be assigned a normal or anomalous label. A case study using a public vehicle trajectory dataset (inD dataset) demonstrates the effectiveness of the proposed method in anomaly detection at four urban intersections.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call