Wearable Gait Dataset With Egocentric Video And Joint Kinematics

327 minutes of synchronized recordings capture 14 walking trials from 10 healthy participants across overground, treadmill, slope, and stair locomotion modes. The dataset combines inertial measurement unit data and shank-mounted egocentric video with high-fidelity ground-truth joint kinematics from optical motion capture. Created by Sy Nguyen and released in April 2026.

Use Cases

• Train models to predict joint angles from IMU signals using synchronized optical motion capture data.
• Develop video-based pose estimation algorithms using 588k egocentric frames per foot paired with ground-truth kinematics.
• Benchmark multimodal fusion techniques for gait analysis by combining time-series IMU data with egocentric video streams.
• Analyze the impact of locomotion mode (treadmill, slope, stair) on sensor data patterns across diverse walking speeds.

Strengths

• 142.7 GB dataset provides 327 minutes of synchronized multimodal recordings
• Includes ground-truth joint kinematics from optical motion capture for 10 participants across 14 trial types
• First known dataset combining wearable IMU and shank-mounted camera data for a wide range of real-world walking tasks

Limitations

• Sample size is limited to 10 healthy participants, which may not generalize to broader or clinical populations
• Geographic and demographic coverage of participants is unknown, potentially introducing bias
• Temporal coverage and recency of the data collection period are not specified

Provenance

Source

Sy Nguyen

Collection Method

Data collected from wearable IMUs, shank-mounted cameras, and synchronized with optical motion capture systems during controlled walking trials.

Freshness

Last updated in April 2026.