TeleRehaB DSS home-based vestibular rehabilitation: wearable sensor recordings.

132,306,879 samples from 11,927 wearable-sensor recordings (199.1 h) of participants in the intervention arm of the TeleRehaB DSS randomised controlled trial of home-based vestibular rehabilitation (ClinicalTrials.gov NCT06534164). Orientation quaternions (and, in walking exercises, three-axis linear acceleration) from head, pelvis and foot sensors, with exercise context and per-recording quality indicators. Anonymised; see README.md for structure, processing and limitations. Participants performed prescribed sitting, standing, walking and stretching exercises at home, guided by a virtual coach, and exergames, over a 12-week programme, wearing up to four Mbientlab MetaMotionS sensors (headband, pelvic belt, both ankles). Orientation comes from on-board NDoF sensor fusion and is reset before each exercise; linear acceleration is in g with gravity removed. All sensors of a recording share a common 10 ms time grid. The dataset consists of three Apache Parquet tables: imu_samples (one row per sample per sensor), recordings (one row per recording and sensor, with duration, sampling and quality statistics) and exercise_events (15,312 exercises from the home-system session logs, with exercise name and level, position in the session, the patient's continue/stop response and the movement metrics computed in real time). datapackage.json documents every column and gives SHA-256 checksums. Coverage is incomplete: 4,260 of 15,045 logged virtual-coach exercises have no uploaded recording, and exercise_events shows which.