watchHAR: Smartwatch IMU and Vicon Motion Data for Daily Activities

Several days of recordings from a Sony Smartwatch 3 and a Vicon motion capture system in the United Kingdom. The dataset contains 3-axial acceleration, gyroscope, and magnetometer data from the smartwatch on both hands, plus ground-truth location data from Vicon for hands and ankles. It covers 18 distinct activities of daily living, including walking variations, personal hygiene, and object manipulation tasks.

Use Cases

• Training activity classification models based on multi-sensor IMU data from a smartwatch.
• Developing sensor fusion techniques by aligning smartwatch IMU streams with high-precision Vicon motion capture data.
• Analyzing fine-grained hand and body kinematics for specific daily tasks like preparing food or washing objects.
• Benchmarking HAR model robustness across different walking styles (freely, with objects, hands in pockets).

Strengths

• Includes synchronized, high-precision ground-truth location data from a Vicon motion capture system for hands and ankles.
• Covers a specific set of 18 realistic activities of daily living, including multiple nuanced walking conditions.
• Contains multi-modal sensor data: 3-axial acceleration, gyroscope, and magnetometer from a commercial smartwatch.

Limitations

• Key metadata such as row count, number of participants, total recording duration, and last update date are unknown.
• Stair activity recordings (stairs_up, stairs_down) lack Vicon ground truth and are only recorded from the dominant (right) hand.
• Data collection was confined to a single room for most activities, which may limit environmental generalization.

Provenance

Source

University of Oxford, author Ada Alevizaki.

Collection Method

Recorded using a Sony Smartwatch 3 (IMU) and a Vicon motion capture system.

Time Range

Recorded across several days (specific dates unknown).

Geography

United Kingdom.