Related Experiment Video
Updated: Sep 16, 2025
16:14
Trajectory Data Analyses for Pedestrian Space-time Activity Study
Published on: February 25, 2013
13.7K
Unsupervised pattern and outlier detection for pedestrian trajectories using diffusion maps
Fanqi Zeng1,2, Nikolai Bode1, Thilo Gross3,4,5
1School of Engineering Mathematics and Technology, University of Bristol, Bristol, BS8 1TW, UK.
Physica A
|July 7, 2025
Summary
Diffusion maps, an unsupervised machine learning method, efficiently analyze pedestrian crowd movement from trajectory data. This approach identifies key dynamics, compares datasets, and detects outliers without prior knowledge.
Area of Science:
- Complex Systems Science
- Computational Social Science
- Data Science
Background:
- Understanding pedestrian crowd dynamics is crucial for real-world applications and fundamental insights into self-driven particle systems.
- Analyzing individual movement paths from trajectory data presents a significant challenge in pedestrian dynamics research.
- Increasing availability of trajectory data necessitates efficient methods for identifying key crowd dynamics features.
Purpose of the Study:
- To demonstrate the utility of diffusion maps, an unsupervised manifold learning technique, for analyzing pedestrian trajectory data.
- To establish an informative feature space for crowd dynamics analysis using trajectory-derived observables.
- To apply diffusion maps to analyze pedestrian movement in diverse scenarios, including a stadium track and room egress.
Main Methods:
- Utilized diffusion maps, an unsupervised manifold learning technique, to analyze pedestrian trajectory data.
- Defined a set of observables from individual movement paths to construct an informative feature space.
- Applied the diffusion map approach to analyze hundreds of trajectories from both stadium track and room egress scenarios.
Main Results:
- Successfully recovered known leading variables governing pedestrian system dynamics using diffusion map analysis.
- Facilitated qualitative comparison of crowd dynamics between experimental and simulated datasets.
- Demonstrated the capability of the approach to automatically detect behavioral outliers within pedestrian trajectories.
Conclusions:
- Diffusion maps offer a computationally efficient, unsupervised method for analyzing pedestrian dynamics from trajectory data.
- The approach requires minimal prior knowledge, making it suitable for live data monitoring and preliminary analysis.
- This method advances the analysis of complex crowd behavior and contributes to the field of pedestrian dynamics research.
Keywords:
Diffusion mapsDimensionality reductionModel validationOutlier detectionPedestrian dynamicsTrajectory analysis
