Spatio-Temporal Vessel Trajectory Clustering Based on Data Mapping and Density

• Main Authors: Huanhuan Li, Jingxian Liu, Kefeng Wu, Zaili Yang, Ryan Wen Liu, Naixue Xiong
• Format: Article
• Language: English
• Published: IEEE 2018-01-01
• Series: IEEE Access
• Subjects: AIS network; data mapping; DBSCAN; trajectory similarity; trajectory clustering; maritime transport
• Online Access: https://ieeexplore.ieee.org/document/8443320/

Automatic identification systems (AISs) serve as a complement to radar systems, and they have been installed and widely used onboard ships to identify targets and improve navigational safety based on a very high-frequency data communication scheme. AIS networks have also been constructed to enhance traffic safety and improve management in main harbors. AISs record vessel trajectories, which include rich traffic flow information, and they represent the foundation for identifying locations and analyzing motion features. However, the inclusion of redundant information will reduce the accuracy of trajectory clustering; therefore, trajectory data mining has become an important research direction. To extract useful information with high accuracy and low computational costs, trajectory mapping and clustering methods are combined in this paper to explore big data acquired from AISs. In particular, the merge distance (MD) is used to measure the similarities between different trajectories, and multidimensional scaling (MDS) is adopted to construct a suitable low-dimensional spatial expression of the similarities between trajectories. An improved density-based spatial clustering of applications with noise (DBSCAN) algorithm is then proposed to cluster spatial points to acquire the optimal cluster. A fusion of the MD, MDS, and improved DBSCAN algorithms can identify the course of trajectories and attain a better clustering performance. Experiments are conducted using a real AIS trajectory database for a bridge area waterway and the Mississippi River to verify the effectiveness of the proposed method. The experiments also show that the newly proposed method presents a higher accuracy than classical ones, such as spectral clustering and affinity propagation clustering.