Havenwatch AI

Home/Nursing homes/How AI fall detection works

How does AI fall detection work in a nursing home?

AI fall detection reads the cameras a home already has and spots a resident who goes to the floor. Care staff get an alert within seconds at any hour. This guide from Havenwatch AI explains each step.

Garden path17:40:00
Two residents walking, no fall
Simulation.

How does AI fall detection work, step by step

The software does five things in a few seconds. A person decides what happens next.

Related: Fall detection camera, Bed exit alert, Night checks.

  1. The camera sends video

    A camera in a corridor, lounge or room sends its picture to a server in the building. The video is read there.

  2. Each person is found

    The software finds each person in every frame. It marks the head, shoulders, hips and knees as points.

  3. The movement is read

    It follows how those points move over a few seconds. A fast drop to the floor, then stillness, looks like a fall.

  4. It checks before it alerts

    A resident who sits down slowly looks different, and so does a staff member who kneels.

  5. Care staff are alerted

    The nurse on duty gets the room and a still. If no one comes, the alert goes to the charge nurse.

What the camera counts as a fall

The software learns the difference from the shape of the movement. Each home can tune it during the pilot.

A fall

A fast drop to the floor is followed by little or no movement. Care staff are alerted at once.

Sitting on the floor on purpose

The resident lowers themselves slowly and with control. That move raises no alert.

A staff member kneeling

A second person bends down beside a resident. That pose raises no alert either.

Lying on a bed or sofa

Lying down on furniture is part of a normal day, so it is left alone.

Why minutes matter after a fall

  • Many older people cannot get up alone after a fall.
  • A call button is often out of reach from the floor.
  • A long time on the floor adds to the harm.
842,000

deaths from falls worldwide in 2023, WHO estimates. Falls are the second-leading cause of unintentional injury death. Source

80%

of people over 90 who fell could not get up after at least one fall, in a Cambridge study. 30% lay on the floor for an hour or more. Source

Public figures from third parties, shown for context. They are not Havenwatch results.

Cameras, wearables and bed mats

Many homes use more than one. Cameras cover the shared spaces where falls also happen.

Pendants and wristbands

They work when worn and pressed. A confused resident may take one off or forget it.

Bed and chair mats

They sense weight leaving one spot. They often ring when a resident only turns over.

Cameras with AI

They need nothing worn or pressed. In rooms they run in privacy view, with consent.

What an alert looks like after a fall

The alert names the room and carries one still in privacy view. Each step is timed, from the fall to help arriving.

Room 12, Level 2Privacy view02:14:06
Room camera in privacy view: a resident lying on the floor beside the bed, drawn as a line outline, the walking frame tipped over
Alert 02:14
Simulation.
Real-time alert02:14:06Closed
A resident is on the floor beside the bed in Room 12.
Location
Room 12, Level 2
Sent to
Care staff on duty, Level 2
Channels
Nurse call, staff phones, Teams
Resolved in
2:10
  1. 02:14Care staff on duty alerted
  2. 02:15Charge nurse, room still empty
  3. 02:16Help reached the resident, seen on camera

Questions nurses ask about AI fall detection

Ask the Havenwatch team a question

How accurate is AI fall detection?

Accuracy depends on the cameras, the light and how a fall is defined. Count hits, misses and false alerts on your own cameras before you buy. The tender checklist shows how.

Does AI fall detection work at night?

Most room cameras switch to infrared at night, and the software reads that picture. Each room camera is checked at night on the walk-through. Any camera to move is named before the pilot.

Does it need cameras in bedrooms?

Rooms are optional at the start. Many homes begin in corridors, lounges and dining rooms, where cameras are already in place. Rooms follow with consent, in privacy view.

Can it predict a fall?

It cannot predict every fall. It can flag the moments that often come first, such as a resident at risk sitting up in bed at night. That is what bed exit alerts do.

Who sees the video?

Video is read on a server in the home. Care staff see one still per alert, and rooms stay in privacy view. Every view of a clip is logged by role.

See AI fall detection on one floor of your home.

A four-week pilot shows how it works on your own cameras, at night as well as by day.