Tracking Human Movement Through Ground Vibrations: Privacy-Preserving Pedestrian Flow Estimation Using a Seismometer

A research team led by graduate student Taewoong Yoon, Project Researcher Ahmad B. Ahmad, and Professor Takeshi Tsuji at the School of Engineering, the University of Tokyo has developed a novel urban monitoring technology that estimates pedestrian flow using only a single seismometer, without relying on cameras, GPS data, or other privacy-sensitive information.

While previous studies have shown that pedestrian-induced ground vibrations can be detected, their performance in noisy urban environments has rarely been quantified, and estimating the number of pedestrians has remained a major challenge. To address this issue, Tsuji-Laboratory developed a deep learning-based method and applied it to real-world seismometer data.

The proposed two-stage framework achieved robust pedestrian detection with an F1-score exceeding 0.92, even in environments affected by traffic and construction noise. Moreover, a regression-based convolutional neural network (CNN) enabled the classification of pedestrian numbers (single individuals, pairs, or groups) with over 80% accuracy.

These results demonstrate that pedestrian flow and congestion can be quantitatively monitored using existing seismometer infrastructure, while preserving individual privacy and avoiding the use of cameras or location-tracking technologies. The technology is expected to contribute to future applications in urban planning, public safety, smart city development, and the unobtrusive monitoring of vulnerable populations such as elderly people and children.

This research was published online on January 5, 2026, in the Springer Nature journal Earth Systems and Environment.

Figure. A two-stage processing framework consisting of pedestrian vibration detection using spectral analysis and pedestrian count classification using a CNN

Papers

Journal: Earth Systems and Environment

Title: Seismometer-Based Pedestrian Monitoring Using Spectral Feature Extraction and Deep Learning: A Privacy-Preserving Approach for Urban Mobility

Authors: Taewoong Yoon, Ahmad B. Ahmad, Takeshi Tsuji*

DOI: 10.1007/s41748-025-01003-4