Uncertainty-aware vessel trajectory prediction for heterogeneous data fusion in internet of things-driven smart waterways

The rapid growth of inland waterway transportation has brought significant challenges in ensuring navigational safety and operational efficiency. Internet of things (IoT)-driven smart waterways systems present innovative solutions to enhance safety, efficiency, and sustainability in inland navigation. Nevertheless, integrating automatic identification system (AIS) and closed-circuit television (CCTV) data remains challenging due to AIS’s low and inconsistent update frequency. It leads to temporal misalignment with real-time video streams, especially in dense or complex scenarios. This work proposes a novel framework to address these challenges with vessel trajectory prediction and heterogeneous data fusion. For trajectory prediction, a spatio-temporal Transformer-based encoder extracts vessel motion features, while a variational autoencoder is employed to capture uncertainties in future trajectories. A gated recurrent unit-driven decoder iteratively predicts high-precision and robust trajectories. The framework aligns AIS-predicted trajectories with CCTV-detected targets using probabilistic matching. It calculates spatio-temporal similarities by combining AIS trajectory distributions with CCTV Gaussian representations. Spatio-temporal similarity computation and optimal assignment-based matching ensure robust and precise correspondences. Extensive experiments on real-world datasets demonstrate that our method improves overall trajectory prediction accuracy by approximately 5.0% compared to the traditional baseline and achieves a matching accuracy of 82.45% in dense traffic scenarios. The proposed framework exhibits excellent robustness under high-density and complex navigation conditions, providing strong support for the practical deployment of IoT-driven smart waterway systems.