AI fall detection
Nobody pressed anything. It gets word out anyway.
A resident on the floor can't reach a pendant. Traditional nurse call produces silence — indistinguishable from someone sleeping peacefully. Foreword sees, asks, and tells you.
Why it matters
The fall isn't the injury. The hour on the floor is.
1 in 4
adults 65+ fall each year, and falls are the leading cause of injury death in that age group.
Source: U.S. Centers for Disease Control and Prevention.
Next round
is how long an unwitnessed fall waits, in a building where nobody's watching. Maybe an hour. Maybe breakfast. The clinical outcome of a fall bends sharply on that number.
90 sec
is what it should be — see it, ask, escalate. Not because the AI is clever, but because it's in the room and it never blinks.
How it works
Pose estimation, on the device, in the room.
A camera feeds a computer vision model running on an edge AI processor at the bedside. The model tracks body position — the geometry of a person, not their face — and recognises the shape of a fall: the angle of the torso, the drop, the stillness that follows.
It watches the room, not the resident
What the model reads is a skeleton: joints and angles. It has no idea what she looks like, and no interest. It's checking whether that geometry just hit the floor.
It asks before it assumes
"The system has detected a potential fall. Are you okay? Say 'I am fine' if you do not need help." A fall detector that just screams is a fall detector staff learn to ignore. This one checks.
Every answer is word
Including no answer. See below — that's the part that matters.
Frames are read and discarded
There's no recording, no upload, no clip in a queue. The model reads the frame and emits an event: a room, a name, a sentence.
Three outcomes
The silence is the loudest one.
I'm fine, I just slipped.
No emergency — but staff still get a record. Falls get logged instead of quietly hidden by an embarrassed resident who'd rather not make a fuss. That's the pattern data your care plans have never had.
I can't get up.
Urgent alert, with her actual words attached. And she heard a voice answer within a second of going down, which is the difference between frightening and terrifying.
This is why the product exists. A fall it saw, followed by a resident who cannot answer, is the most important thing anyone could tell you — and it's exactly the thing a pendant on a bedside table will never say.
Privacy
She can switch the camera view off. She's still caught.
Five privacy levels, chosen by the resident from her own phone. Staff can see live video — or a stick-figure outline and nothing else — or no video at all. Every one of them still detects falls.
That isn't a compromise we engineered around. It falls out of the architecture: the model is in the room, so the image was never going anywhere. "Camera view off" simply means staff don't see it either. Every other approach makes you trade privacy against safety, because their intelligence lives somewhere the video has to travel to.
Honest about it
What we don't claim.
It isn't a medical device
Foreword detects a fall and tells your staff. It doesn't diagnose, it doesn't triage clinically, and it doesn't replace assessment by someone qualified to do it.
It isn't perfect, and we'd distrust anyone who said theirs was
Vision systems have false positives and edge cases. Ours is tuned to over-call rather than under-call: a caregiver checking on someone who's fine is a cheap error. The other kind isn't.
Watch it catch one.
We'll show you a real detection on real hardware — including what your staff see.
Book a demo