CAPe: A Continuous and Adaptive Framework for Trajectory Co-movement Pattern Detection

Abstract

Co-movement pattern detection aims to identify spatiotemporally and semantically correlated movement behaviors among large groups of moving objects, and it serves as a core analytical primitive in trajectory-based applications such as anomaly detection and trend forecasting. Existing approaches primarily rely on clustering over timestamp-based snapshots, followed by indexing and optimization techniques to boost efficiency. However, these snapshot-driven designs capture only discrete spatial states, failing to preserve continuous movement dynamics and remaining highly vulnerable to sampling misalignment in trajectory streams. Furthermore, although learning-based models can incorporate richer behavioral features, their intrinsic training and inference overhead limit applicability in continuous, real-time settings. This work introduces CAPe, a Continuous and Adaptive framework for co-movement Pattern detection designed for streaming trajectories with unknown and heterogeneous sampling rates. Specifically, CAPe encompasses three modules: segment-based snapshot clipping, sub-trajectory clustering and pattern enumeration. (i) To preserve movement continuity, a segment-based snapshot mechanism is introduced, with a theoretical proof of θ-completeness provided to relax the boundary assumptions. (ii) For efficient and accurate sub-trajectory clustering, SKD is proposed as a lightweight kNN-DBSCAN variant tailored to segment data. To improve scalability, sub-trajectory similarity computation is optimized through a series of DTWacceleration techniques. (iii) For large-scale pattern enumeration, index-based partitioning on Flink is adopted, and traditional co-movement constraints are generalized to operate over segment-based snapshots. Extensive experiments on real-world datasets demonstrate that CAPe consistently outperforms state-of-the-art methods in both efficiency and effectiveness, highlighting its potential as a continuous and adaptive paradigm for trajectory co-movement pattern detection.

Keywords

Co-movement Pattern Detection / Trajectory Streaming / Real-time Processing / Spatiotemporal Data Mining / Sub-trajectory Clustering

RIGHTS & PERMISSIONS

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.