Sensor & IMU → Activity recognition (sensor)
Sensor & IMU for Activity recognition (sensor)
Activity recognition (sensor) needs labelled IMU/sensor streams across activities and devices. Here is why Sensor & IMU is the right raw material for it, what buyers typically spec, and how the rights are handled.
Why Sensor & IMU for Activity recognition (sensor)
Activity recognition from wearables is a label-quality problem wearing a data-volume costume. IMU streams are cheap; trustworthy ground truth about what the wearer was doing is not, and every corpus lives or dies on how labels were verified — video-verified beats observer-logged beats self-reported, and the method belongs in the spec. The second axis is placement and device: accelerometer signatures differ enough between wrist, pocket, and chest that models transfer poorly across placements, so the corpus must match deployment or cover placements explicitly. Third is population: gait and movement vary by age, body type, and mobility, and a corpus of university volunteers quietly fails on the users many products target. Scripted sessions give dense, clean labels; free-living collection gives honest transitions and ambiguity. Production briefs almost always want a scripted core plus a free-living evaluation slice.
What buyers typically spec
Industry-typical ranges — a brief can and should deviate where the task demands it.
| Typical volume | Hundreds of participant-hours; participant count weighted over hours-per-participant |
|---|---|
| Signals | 6–9-axis IMU at 50–200 Hz; device model and placement per session |
| Labels | Activity spans with boundaries; verification method stated (video/observer/self-report) |
| Coverage | Placement variety, device variety, population diversity, class balance report |
| Formats | CSV/JSON streams with one timebase |
A sample brief
The shape of a workable request — swap in your own numbers and conditions:
- Modality: wrist and pocket IMU during scripted activities plus free-living days.
- Volume: 200 participants; 2 h scripted each + 8 h free-living subset.
- Labels: video-verified spans for scripted sessions; diary-verified for free-living.
- Rights: participant consent naming AI training; motion data treated as personal data.
fiund's sourcing angle
Sensor data is abundant on devices but almost never licensed cleanly. fiund sources consented streams to spec. We source to a brief and clear the rights before anything moves, so what you receive is both useful and defensible in diligence.
Rights posture
Signed licence, explicit training rights, separate voice/likeness consent, nothing scraped. See the rights & provenance guides.
Frequently asked questions
Which label verification method is worth paying for?
Video verification is the ceiling — an annotator can check exactly when an activity started. Observer logging is next; unverified self-report drifts by minutes and mislabels transitions. Price scales in the same order, and the scripted/free-living split lets you spend where it counts.
Will a wrist-trained model work on phone-in-pocket data?
Poorly, out of the box — placement changes the signal geometry and dominant frequencies. Either collect the placements you will deploy on, or spec multi-placement sessions so the same activities exist across positions for transfer work.
More Sensor & IMU use cases
Need Sensor & IMU for Activity recognition (sensor)?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief