16 Paper Presentations at ICDCN2026
16 presentations and 2 awards at ICDCN 2026 held in Nara, Japan (January 2026)
Publications & Talks Awards
International Conference · 2026
Short-Term Crowd Prediction through Continuous Observation
Abstract
In this study, we propose a data assimilation–based crowd digital twin framework that integrates real-time sensing data with an agent-based simulation. The proposed system continuously updates virtual agents in a 3D building model based on pedestrian positions obtained from LiDAR sensors, enabling both short-term and long-term predictions of crowd dynamics. To correct the temporal discrepancy between real-world observations and simulated virtual time, we introduce a Lead–Chase framework: the Lead simulation performs high-fidelity future prediction, while the Chase simulation dynamically adjusts the results using the latest observations. Evaluation using real pedestrian tracking data collected by the LiDAR-based system installed on the campus building demonstrated that the Lead–Chase framework approach reduced the average displacement error by approximately 50% compared to a single simulation, while maintaining computation time suitable for real-time processing.
Research Note
Predicting how a crowd will move over the next few tens of seconds is what turns crowd sensing into crowd management, because staff can only act on information that arrives before the congestion does. The difficulty lies in the mismatch between observation and simulation. A simulation that predicts the future is by definition running ahead of real time, so by the moment a fresh sensor reading arrives the simulated world has already moved on.
Data assimilation is the established way of handling this. Long used in weather forecasting, it keeps a running simulation anchored to reality by feeding measurements into its internal state rather than restarting it from scratch. Applied to pedestrian dynamics, it means continuously correcting virtual agents so that they follow the people the sensors actually see.
This work builds such a crowd digital twin on top of an agent-based simulation and a 3D building model, driven by pedestrian positions from LiDAR sensors. The temporal gap is handled by running two coupled simulations. One of them looks ahead and produces a high-fidelity forecast, while the other adjusts that forecast using the most recent observations.
Evaluated on real pedestrian tracking data collected in a campus building, the approach roughly halved the average displacement error compared with a single simulation while staying within a computation budget suitable for real-time use.
Related Lab Activities
Related Projects
A Platform for Digitalizing Knowledge of Regional Communities
Japan Science and Technology Agency (JST) CREST
「基礎理論とシステム基盤技術の融合によるSociety 5.0のための基盤ソフトウェアの創出」領域
LLM-Driven Urban Transportation Simulation Platform
Japan Science and Technology Agency (JST) JST Strategic Basic Research Programs (PRESTO)
[Social Transformation Platform] Co-Creation of the Transformation Platform Technology for Human and Society by Integration of the Humanities and Sciences
