Reliable trajectory similarity estimation underpins applications such as map matching, mobility anomaly detection, and travel behavior analysis. However, traditional distance-based methods are computationally expensive and sensitive to irregular sampling, while learning-based approaches, though more efficient, often neglect geometric consistency and degrade under sparse or noisy trajectory data. This paper presents GeoCPC-TrajSim, a self-supervised framework for trajectory similarity estimation that employs an Information Noise-Contrastive Estimation (InfoNCE) objective aligned with shape geometry through a shape-consistent positive–negative sampling strategy. To evaluate and preserve this alignment, we establish the Trajectory Geometry-Consistency Benchmark (TGCB), a unified metric space for both optimization and assessment based on Fréchet, Hausdorff, and directional dispersion measures. A Spatio-Temporal Augmentation and Alignment Module (SAAM) introduces two augmentations: sliding-window interpolation, which densifies sparse trajectories while maintaining curvature continuity, and detour offset injection, which simulates realistic Global Navigation Satellite System (GNSS) drift. A Multi-Factor Feature Extraction Module (MFEM) enriches the representation with nine kinematic and directional descriptors beyond conventional spatial–temporal inputs, mitigating the information loss inherent in sparse trajectories. Experiments on Grab-Posisi and GeoLife datasets demonstrate that GeoCPC-TrajSim improves accuracy, robustness, and efficiency in trajectory similarity estimation while reducing training cost by over 70% compared with recent baselines.

Urban traffic flow forecasting remains challenging under heterogeneous urban functions, dynamic environmental conditions, and temporary gaps in recent regional observations. Existing spatio-temporal models primarily rely on historical traffic flows and often lack explicit representations of regional structure and urban semantics. We propose VLM-enhanced Context-aware Prior-guided Prediction with Decoupled Semantic Refinement (VLM-CPPD), a multimodal framework that integrates traffic dynamics with urban morphology, environmental context, and vision-language semantic priors derived from remote-sensing imagery. We further construct Porto-MMTF, a grid-aligned multimodal dataset that integrates inflow/outflow sequences, weather conditions, temporal context, OpenStreetMap morphology, remote-sensing imagery, and structured semantics extracted offline by MM-RSTraj. The core framework combines morphology-conditioned environmental adaptation, peak-aware forecast refinement, and a lightweight semantic-refinement interface. MM-RSTraj is used exclusively for offline semantic construction and is fully decoupled from online forecasting. Beyond the core framework, we introduce a masked-history extension for regions whose recent traffic observations are temporarily unavailable. In this setting, target-region structural and semantic information is combined with current traffic dynamics from observed grids to support prediction. On Porto-MMTF, VLM-CPPD achieves an RMSE of 3.7485, an MAE of 1.5435, and a Top10 MAE of 6.3692. On TaxiNYC, after city-specific data alignment and retraining, VLM-CPPD reduces RMSE, MAE, and Top10 MAE by 6.30%, 4.99%, and 6.08%, respectively, relative to STDN.

Trajectory traffic semantic understanding is fundamental to applications such as intelligent transportation and urban mobility analysis. While multimodal large language models (MLLMs) have recently advanced remote sensing scene understanding, current models remain focused on general remote sensing semantics and lack tailored designs for trajectory-specific tasks. To bridge this gap, we propose MM-RSTraj, the first remote sensing–assisted multimodal framework tailored for trajectory traffic semantic understanding. Built upon the LLaVA-OneVision architecture, MM-RSTraj adopts a two-stage fine-tuning strategy to enhance cross-modal interaction between remote sensing imagery and trajectory features. To support this process, we construct two high-quality instruction datasets: RSI-Instruct, an extension of RSICap providing multi-turn instruction–response supervision for general remote sensing semantics; and RSI-Traffic, a dataset designed for trajectory traffic semantic understanding, emphasizing key environmental semantics such as road structures, building layouts, and trajectory-related features. Experimental results demonstrate that MM-RSTraj achieves superior performance in remote sensing trajectory traffic semantic evaluation, while also attaining competitive results in general remote sensing semantic tasks such as RSIC and RSVQA. This work establishes a new paradigm for integrating environmental semantics with trajectory modeling through multimodal large language models (MLLMs).