Anomalous Trajectory Detection Based on Multi Sector Division Feature Representation

Abstrak

Trajectory anomaly detection plays a crucial role in cloud–edge collaborative intelligent border security systems as one of the key data mining technologies. However, existing detection methods are often constrained by inherent detector design limitations and insufficient trajectory feature representation capabilities. To overcome these challenges, this paper introduces a trajectory anomaly detection algorithm based on Multi-Sector Division Feature Representation (MSDFR). The proposed method first partitions the longitude-latitude plane into equal-angle sectors to process variable-length trajectory data. It subsequently extracts discriminative trajectory features by measuring the distance between each trajectory point and a central reference point within the corresponding sector, in addition to incorporating polar coordinate angular offsets. An unsupervised learning algorithm is then employed to identify anomalous trajectories based on these enhanced features. Experimental results demonstrate that the proposed MSDFR method can effectively detect anomalous trajectories, and the extracted features exhibit promising generalization potential for a variety of trajectory mining applications.

Identifier

• OpenAlex: https://openalex.org/W7165205352
• DOI: https://doi.org/10.1109/access.2026.3705006