VIBEID: A STRUCTURAL VIBRATION-BASED SOFT BIOMETRIC DATASET FOR HUMAN GAIT RECOGNITION

Mainak Chakraborty, Chandan, Sahil Anchal, Bodhibrata Mukhopadhyay, Subrat Kar

Keywords: Structural vibrations, Gait Recognition, Deep learning, Machine learning

TL;DR: VIBeID offers a dataset and benchmark for person identification using structural vibrations from walking on three surfaces and three distances from sensor, enabling comparison with video-based methods.

Abstract:

We present VIBeID, a dataset and benchmark designed for advancing non-invasive human gait recognition using structural vibration. Structural vibrations, produced by the rhythmic impact of the toe and heel on the ground, are distinct and can be used as a privacy-preserving and non-cooperative soft-biometric modality. We curated the largest dataset VIBeID consists of footfall generated structural vibrations of 100 subjects. Existing datasets in this field typically include around ten subjects and lack comprehensive exploration of domain adaptation. To thoroughly explore the domain adaptation aspect of this biometric approach, we recorded vibration data on three distinct floor types (wooden, carpet, and cement) and at three distances from the geophone sensor (1.5 m, 2.5 m, and 4.0 m), involving 40
and 30 subjects, respectively. Additionally, we benchmarked our dataset against video recordings from 15 individuals in an outdoor setting. Beyond providing 88 hours of raw vibration data, VIBeID establishes a comprehensive benchmark for a) person identification: where the aim is to recognize individuals through their unique structural vibrations, b) domain adaptation: assessing model performance across different walking surfaces and sensor positions, and c) multi-modal comparison: comparing vibration-based and vision-based identification methods. Our experiments, using both machine learning and deep learning approaches, establish a baseline for future research in this field, and introduce a large-scale dataset for the broader machine learning community.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10133