TOWalk: A Multi-Modal Dataset for Real-World Movement Analysis

The TOWalk Dataset has been developed to support research on gait analysis, with a focus on leveraging data from head-worn sensors combined with other wearable devices. This dataset provides an extensive collection of movement data captured in both controlled laboratory settings and natural, unsupervised real-world conditions in Turin (Italy).

The dataset features high-quality recordings from 15 healthy young participants, using the INDIP system alongside head-worn and wrist-mounted inertial sensors. The INDIP system includes two foot-mounted MIMUs, pressure-sensing insoles, and two ankle-mounted time-of-flight distance sensors. Additional head-worn and wrist-mounted MIMUs complement this setup, offering comprehensive insights into human locomotion.

The TOWalk Dataset has already yielded excellent results in studies involving gait speed estimation, real-world gait event segmentation, and gait sequence detection, proving its utility for researchers working on wearable-based gait analysis and mobility monitoring in diverse settings.

Key Features

• Subjects: 15 healthy young participants.
• Devices: INDIP reference system (two foot-mounted MIMUs, one lower-back MIMU, two foot pressure insoles, and two ankle-mounted distance sensors), one head-worn MIMU, and two wrist-worn MIMUs.
• Acquisition Conditions: Lab-based tests and unsupervised real-world activities.
• Data Types: Raw sensor data, segmented gait sequences, strides, and spatial-temporal gait parameters.

Lab-Based Acquisitions

Lab-based data were recorded in controlled indoor conditions through the following structured tests:

1. Static Acquisition: Sensors placed on a flat surface, data collected in static conditions.
2. Standing Acquisition: Subjects standing still while wearing sensors.
3. Data Personalization: Controlled tasks, including standing, raising arms and legs, and walking 12 meters at a comfortable speed, to verify data quality.
4. Walking Straight at Slow Speed: 12-meter walk at a slow speed (three trials).
5. Walking Straight at Comfortable Speed: 12-meter walk at a comfortable speed (three trials).
6. Walking Straight at Fast Speed: 12-meter walk at a fast speed (three trials).
7. Walking with Turns: Walking back and forth over 12 meters with turns (three trials).

Tests 1–3 are non-walking and primarily used for calibration. When performing gait-related operations, these tests should be excluded.

Each trial began and ended with the participant standing still.

Real-World Acquisitions

Real-world data were collected under natural, unsupervised conditions to ensure ecological validity. Participants were not given specific instructions but were required to:

• Raise and sit from a chair at least once.
• Climb stairs or ramps at least once.
• Enter a new room at least once.

This setup ensured the inclusion of diverse, unconstrained movements.

For real-world acquisitions, Recording4 contains the actual activity data. Recordings 1–3 are calibration recordings and are not required for most analyses.

Data Organization

The dataset is structured by subject and condition:

• /data.mat: Raw sensor data sorted by subject ID and acquisition condition ("Lab-based" or "Real-world").
• /participants_summary.xlsx: Demographic information for each participant.

Files for Each Recording

• infoForAlgo.mat: Metadata containing demographic and sensor-related details.
• data.mat: Raw sensor data, spatial-temporal gait parameters, and metadata from the INDIP system.

Notes and Recommendations

Data Quality

All recordings have undergone rigorous quality checks to ensure reliability.

General Notes

• Participant IDs are pseudonymized for privacy.
• The dataset is intended for research and algorithm validation, not clinical applications.

Suggested Citation

If you use the TOWalk Dataset in your research, please cite the following works:

1. Salis, F., Bertuletti, S., Bonci, T., et al. (2023). “A Multi-Sensor Wearable System for the Assessment of Diseased Gait in Real-World Conditions.” Frontiers in Bioengineering and Biotechnology, 11:1143248. DOI: 10.3389/fbioe.2023.1143248.
2. Tasca, P., et al. (2023). "A machine learning-based pipeline for stride speed estimation with a head-worn inertial sensor." Gait & Posture. DOI: 10.1016/j.gaitpost.2023.07.341.
3. Tasca, P., et al. (2023). "A machine learning approach for stride speed estimation based on a head-mounted IMU." GNB2023 Conference Proceedings. ResearchGate Link.
4. Tasca, P., et al. (2024). "Estimating Gait Events and Speed in the Real World with a Head-Worn IMU." (pre-print). DOI: 10.36227/techrxiv.170654480.02767120/v1.
5. Tasca, P., et al. (2024). "Real-world gait detection with a head-worn inertial unit and features-based machine learning." Gait & Posture. DOI: 10.1016/j.gaitpost.2024.08.068.

License and Disclaimer

The TOWalk Dataset © 2025 is licensed under CC BY-NC-ND 4.0. It is provided exclusively for research purposes. The authors make no guarantees regarding the dataset's accuracy or reliability and are not responsible for any misuse or misinterpretation of the data.