A Structural Vibration-based Dataset for Human Gait Recognition.

We present a dataset designed to advance non-intrusive human gait recognition using structural vibration. Structural vibrations, resulting from the rhythmic impacts of toes and heels on the ground, offer a unique, privacy-preserving gait recognition modality. We curated the largest dataset consisting of structural vibration signals from 100 subjects. Existing datasets in this domain are limited in scope, typically involving around ten participants and offering minimal exploration. To comprehensively investigate this modality, we recorded vibration signals across three distinct floor types-wooden, carpet, and cement-and at three different distances from a geophone sensor (1.5 m, 2.5 m, and 4.0 m), involving 40 and 30 participants, respectively. The dataset also includes video recordings of 15 individuals in an outdoor setting. Moreover, we recorded structural vibration signals of 15 people walking at three different speeds. Alongside the vibration data, we provide physiological details such as participant age, gender, height, and weight. The dataset contains over 96 hours of raw structural vibration data, along with additional interim and processed data. This dataset aims to address long-standing challenges in non-intrusive and privacy-preserving gait recognition, with potential applications in clinical analysis, elderly care and rehabilitation engineering.