SOSC is a deployed sensing infrastructure that instruments everyday objects with smart tags and integrates commercial smartwatch data to capture how users interact with Objects of Daily Living. The platform has been operational at Helse Campus Stavanger since 2023, collecting synchronized multimodal data without requiring manual annotation.
SOSC has been designed for ecological validity: data is collected in real home and care environments without constraining user behaviour. The platform has been deployed and validated at HCS, where it supports multimodal data collection with elderly participants in smart home settings.
SOSC has been designed and validated with off-the-shelf commercial smartwatches running Wear OS. The platform's data acquisition layer is compatible with the following devices, making deployment accessible without specialised hardware.
All listed devices expose IMU (accelerometer, gyroscope, magnetometer), heart rate, GPS, and BLE scanning APIs via the Wear OS Health Services and Sensors API, which SOSC uses for data acquisition. NFC tag reading is supported on all devices through the Android NFC stack.
Each sensing layer contributes distinct signal types that are collected, timestamped, and stored in a synchronised record. The table below summarises the signal types collected by the platform and their role in the interaction recognition pipeline.
| Signal | Source | Variables | Role in pipeline | Type |
|---|---|---|---|---|
| Wrist acceleration | Smartwatch IMU | 3-axis, up to 200 Hz | Primary motion feature for interaction segmentation and activity classification | motion |
| Gyroscope / magnetometer | Smartwatch IMU | 6-axis, rotation matrix | Wrist orientation during object manipulation; activity transition detection | motion |
| Heart rate / HRV | Optical PPG | HR bpm, beat-to-beat interval | Physiological state; auxiliary feature for disambiguating similar ADL profiles | physiology |
| GPS / indoor position | GPS + BLE beacons | Outdoor coordinates, indoor proximity zone | Spatial context; room-level localisation supporting environment-driven labelling | spatial |
| Object proximity / contact | Smart tags — BLE/NFC | Object ID, RSSI, contact timestamp | Core signal for interaction-driven label inference; object identity and usage sequence | interaction |
| Temperature / pressure / light | Fixed environmental sensors | °C, hPa, lux | Ambient context; supports disambiguation of activities sharing similar motion profiles | environment |
The core capability of SOSC is the inference of activity labels from structured sequences of human–object interaction events, without requiring manual annotation. The pipeline maps raw tag contact events and wrist motion patterns to semantic ADL labels via a two-stage mechanism validated on platform data.
SOSC has been validated through a series of data collection studies at HCS. The results demonstrate the feasibility of interaction-driven activity recognition from commercial wearable devices and smart tags in ecologically valid settings.