Fine-Grained Ship Trajectory Multi-Anomaly Integrated Detection Based on Trajectory Clustering

Aiming at the anomaly detection of ship trajectory, this paper proposes a fine-grained multi-anomaly integrated detection method by taking the trajectory clustering as the basis of the normal model establishment, and taking the ship trajectory of the local sea area as the object of analysis. The normal model is established with the track segment as the basic unit, and the track segment is firstly extracted by using the track segment clustering method, and then the 2D single Gaussian model, Gaussian hybrid model and heading discretization model are used to complete the modeling of the ship's positional distribution, speed distribution and heading distribution in the track segment, so as to complete the detection of positional anomalies, cross-harbor anomalies and behavioral anomalies. The experimental results prove the effectiveness of the detection method.