CAPTURE-24: A large dataset of wrist-worn activity tracker data collected in the wild for human activity recognition

Summary

CAPTURE-24 is a large-scale, free-living wrist accelerometer dataset released by Oxford's OxWearables group, designed to enable reproducible benchmarking of human activity recognition (HAR) algorithms for wrist-worn devices. Collected from 151 participants each wearing an Axivity AX3 wrist accelerometer (100 Hz, ±8g) for ~24 hours in unmonitored real-world conditions, the dataset yields 3,883 hours of raw triaxial accelerometer data, of which 2,562 hours are annotated with activity labels derived from body-worn cameras and sleep diaries. This makes CAPTURE-24 two to three orders of magnitude larger than any prior publicly available annotated HAR dataset.