Distributed acoustic sensing enables real-time traffic monitoring and vehicle motion-state prediction

Distributed optical fiber sensing enables infrastructure-friendly, continuous, and real-time observation of dynamic environments. Among these technologies, Distributed Acoustic Sensing (DAS) provides a promising optical sensing modality for capturing traffic-induced vibrations along fiber routes. This study proposes a real-time traffic monitoring and vehicle motion-state prediction framework based on DAS. We use YOLOv11 to achieve efficient vehicle detection and DAS-derived speed estimation, accurately distinguishing closely spaced vehicles. Based on the reconstructed observations, we further introduce a Trend Modeling + Deep Kernel Learning (Trend+DKL) framework to reconstruct and predict multi-vehicle trajectories, integrating motion trends with nonlinear uncertainty modeling. Experiments in real-world scenarios demonstrate that the proposed method effectively supports multiple tasks, including speed estimation, traffic counting, and motion-state modeling. The vehicle detection rate exceeds 96.7%, with an average speed-estimation MAE of 1.45 km/h and a motion-state prediction MAE as low as 0.95 m/s. These results highlight the practical potential of DAS for intelligent traffic monitoring and prediction, offering new perspectives for future intelligent transportation systems.