Key Route Node Extraction from AIS Trajectories via Multi-Constraint Turning Point Identification and Heading-Aware Adaptive DBSCAN

Automatic Identification System (AIS) trajectories provide valuable spatiotemporal information for maritime route structure mining, but robust extraction of key route nodes remains difficult because raw data are noisy, turning behaviors are easily masked by local fluctuations, and conventional Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is sensitive to fixed parameters and ignores heading differences. To address these issues, this study proposes a key route node extraction framework based on multi-constraint turning-point identification and heading-aware adaptive DBSCAN (HA-DBSCAN). Raw AIS data are first cleaned, segmented, and compressed using a heading-aware Douglas–Peucker strategy to reduce redundancy while preserving geometric and directional characteristics. Valid turning points are then identified by jointly considering heading change rate, geometric curvature, and temporal stability. Finally, HA-DBSCAN integrates a heading-aware distance metric, adaptive neighborhood estimation, and density-aware MinPts optimization to cluster turning points and extract representative route nodes. Experiments using AIS data from the Ningbo–Zhoushan Port area retained 287,614 valid records and 754 continuous trajectory segments, from which 1710 turning points were identified. The proposed method generated 45 stable clusters with a noise ratio of 0.0450 and route coverage of 95.5%. These results indicate that, within the current study setting, the framework can distinguish crossing routes, adapt to heterogeneous traffic densities, and provide an interpretable intermediate layer for subsequent maritime route-structure modeling. Supplementary validation on the same AIS dataset further showed that, compared with DBSCAN, Ordering Points To Identify the Clustering Structure (OPTICS), and HDBSCAN baselines as well as several pipeline ablations, the full framework achieved a more balanced performance in terms of coverage, noise suppression, and avoidance of cluster over-fragmentation.