Real-Time Transportation Origin-Destination Demand Estimation Using Multimodality Data

Term Start:

July 1, 2025

Term End:

May 31, 2026

Budget:

$60,000

Keywords:

Multimodal OD Estimation, Real-Time Mobility Modeling, Trajectory–Text Data Fusion

Thrust Area(s):

Data Modeling and Analytic Tools

University Lead:

University of Washington

Researcher(s):

Xuegang (Jeff) Ban

Real-time origin-destination (OD) demand estimation is essential for urban transportation management, enabling responsive traffic control, dynamic transit scheduling, e-hailing services, and emergency operations. Traditional OD estimation methods rely heavily on traffic sensor data and statistical modeling; but the limited coverage of sensors in most cities poses a significant challenge. Emerging data sources—such as mobile trajectories from smartphones and connected vehicles, and textual traffic information from reports or social media—offer promising alternatives for capturing real-time mobility patterns at scale and lower cost. However, current approaches often fail to integrate these heterogeneous data sources into a unified framework.

This project proposes a multimodal learning framework that fuses traffic time series data, mobile trajectory data, and transportation-related textual information to improve real-time OD demand estimation. Static OD matrices serve as a prior, which are dynamically adjusted based on real-time multimodal inputs. We leverage time series models for traffic counts, sequence and graph-based methods for trajectory data, and large language models to extract OD-relevant insights from text. Additionally, we aim to construct a hybrid dataset combining traffic counts, mobile trajectories, and text, derived from simulation or real-world sources, to support model development and benchmarking. Our work seeks to advance robust, scalable, and intelligent mobility solutions.

By integrating multimodal data sources in real-time OD estimation, our framework not only estimates OD demands but also provides deeper insights into travel behavior, particularly in destination and path selection. The fusion of trajectory patterns with contextual textual cues enables us to infer why travelers choose certain destinations—whether due to events, service availability, or disruptions. Similarly, sequence-based modeling of trajectories reveals preferred routes, shedding light on dynamic path choices under varying traffic conditions. This enriched understanding supports the development of more adaptive and human-centric transportation systems, informing planning decisions, optimizing route recommendations, and enhancing real-time traffic interventions based on actual travel motivations and behaviors.

Project scope

Scroll to Top