The FootprintID Dataset: Footstep-Induced Structural Vibration Data for Indoor Person Identification with Different Walking Speeds

The dataset consists of structural vibration data (vertical velocity of floor structure) induced by 10 people’s footsteps as they walk around with 8 different walking speeds, sensed by 5 geophone sensors.

The footstep-induced structural vibration data is stored as footstep traces, each consisting of a series of consecutive footsteps (see the sample plot). The dataset is stored in a MAT-file named People.mat. The dataset includes three layers of labels: 1) person identity i (i = 1, 2, ..., 10), 2) sensor number j (j = 1, 2, ..., 5), and 3) walking speed k (k = 1, 2, ..., 8). The speed k represents the walking speeds of \(\mu,\ \ \mu+\sigma,\ \ \mu+2\sigma,\ \ \mu+3\sigma,\ \ \mu-\sigma,\ \ \mu-2\sigma,\ \ \mu-3\sigma\), and self-selected speed by each person respectively. \(\mu\) and \(\sigma\) refer to the mean and standard deviation of the step frequencies. To access the footstep traces from the person i, sensor j with walking speed k, please use the MATLAB syntax People{i}.Sen{j}.S{k}. This gives a \(m\times n\) cell structure. \(m\) denotes the individual trace number, of which the number of traces varies from 10 to 12; \(n\) represents the level of amplification, including 2000X, 4000X, and 6000X, corresponding to n = 1, n = 2, and n = 3 respectively. To read and plot a sample trace of footstep-induced floor vibration, use the script read_data.m. For more details, please refer to the original FootprintID paper in the following link: https://dl-acm-org.stanford.idm.oclc.org/doi/10.1145/3130954

The human walking experiment involves 10 participants aged between 20 to 29 years old, of which 8 are male and 2 are female. Their walking area is 30ft X 6ft along a hallway with concrete floor. Each of the participants wears flat bottom shoes.

The sensing unit consists of 5 components: 1) the geophone (SM-24), 2) the amplification module, 3) the processor board, 4) the communication module (XBee radio), and 5) the batteries. The sensing unit converts the structural vibration velocity into voltages records. The sampling frequency is 1000Hz.

The hardware unit, experiment setup, and a sample data plot can be found in Experiment Introduction.pdf. Further implementation details can be found in the original FootPrintID paper in the link above.

Please cite this dataset as:

Yiwen Dong, Shijia Pan, Tong Yu, Mostafa Mirshekari, Jonathon Fagert, Amelie Bonde, Ole J. Mengshoel, Pei Zhang, and Hae Young Noh. 2021. The FootprintID Dataset: Footstep-Induced Structural Vibration Data for Person Identification with 8 Different Walking Speeds. Zenodo, DOI: https://doi.org/10.5281/zenodo.4691144

Shijia Pan, Tong Yu, Mostafa Mirshekari, Jonathon Fagert, Amelie Bonde, Ole J. Mengshoel, Hae Young Noh, and Pei Zhang. 2017. FootprintID: Indoor Pedestrian Identification through Ambient Structural Vibration Sensing. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 1, 3, Article 89 (September 2017), 31 pages. DOI: https://doi.org/10.1145/3130954