Last updated October 2026
Elderly fall prevention and fall detection in care homes
A fall detection system reads a room or bed continuously and raises a possible-fall alert when posture drops toward the floor, or stops changing for too long. It is a prompt for staff to check, not a diagnosis, and it never replaces the judgement of the person who arrives.
Fall detection exists to tell the staff on shift, quickly, that something may have happened in a room — not to replace their judgement. SteadiCore SI (artificial intelligence AI) reads Wi-Fi signal disturbance to estimate posture in the room (an RF Pose Estimate), raises a possible fall alert when the pattern matches a fall-like event, and requires staff to verify it. This page explains what such a system actually measures, how it differs from cameras and wearables, and the Hong Kong compliance context a care home should know before it buys one.
What a fall detection system is
A fall detection system is an alerting layer. It reads a signal continuously at room or bed level, and when the shape of that signal matches a fall — or matches prolonged stillness — it raises an alert that a person has to act on. It does not diagnose, it does not keep a medical record, and it does not decide for staff whether someone actually fell.
In the daily running of a care home the problem it addresses is a gap in time: the gap between an event and somebody knowing about it. That gap widens between night rounds, behind a closed bathroom door, and whenever one carer is looking after several residents at once. The point of the system is to narrow the gap and tell staff which bed or which bathroom to go to.
What it does and does not do
- Does: monitor posture change in a room continuously and alert on a fall-like event.
- Does: flag prolonged stillness, such as a resident who has left the bed and not returned.
- Does: log every alert, arrival time and outcome, so the sequence can be reviewed afterwards and reported on.
- Does not: guarantee that every fall is detected. Every sensing method has blind spots, and the output is always an alert that requires verification.
- Does not: identify residents by facial recognition, and does not put a camera in a bedroom.
- Does not: give a medical diagnosis, medication advice, or a clinical conclusion about fall risk.
How a Wi-Fi fall sensor works
A Wi-Fi fall sensor needs no camera and nothing worn by the resident. A transmitter keeps sending Wi-Fi signal into the room and a receiver reads how that signal is disturbed; a person standing, walking or lying down changes the shape of those disturbances. The system turns the disturbance into a posture estimate for the room (RF Pose Estimate), then asks whether the posture sequence matches a fall-like event. What leaves the room is a posture estimate and an event, not an image.
The four stages, signal to prompt
- Signal — a transmitter and receiver keep a Wi-Fi link open; a body moving through it produces measurable disturbance.
- Posture estimate — the disturbance is converted into a posture for the room and tracked over time.
- Event decision — posture dropping to the floor inside a very short window, or barely changing across a long one, produces an event.
- Verification and record — the event reaches the care station unverified, and staff record what they found.
What a fall-like pattern and prolonged stillness mean
A fall-like pattern is a posture that moves from standing or sitting to near floor level within a very short window, and does not return to the earlier posture. What the system sees is that shape, not the conclusion "a person has fallen" — a member of staff has to confirm the rest, which is why the wording is a possible fall alert.
Prolonged stillness means the posture estimate in a room barely changes for a long stretch during a period when movement would be expected. It may be rest and it may be a situation that needs help. The system's job is to surface the state; the judgement stays with the staff on shift.
Wi-Fi sensing compared with cameras and wearables
All three can raise an alert. They differ in what they can see and in the conditions where they stop working. The decision is usually less about an accuracy figure than about which trade-off a home is willing to accept: privacy, compliance from the resident, or installation work.
Trade-offs between the three common sensing methods
What it measures
- Wi-Fi sensing
- Disturbance of a signal in the room, turned into a posture estimate
- Cameras
- Image
- Wearables
- Acceleration and orientation of the device
Bedroom privacy
- Wi-Fi sensing
- No camera in the room; only a posture estimate is produced
- Cameras
- Requires images, usually with masking or blurring
- Wearables
- No images involved
Resident cooperation
- Wi-Fi sensing
- None required — nothing is worn
- Cameras
- None to wear, but consent to being filmed
- Wearables
- Daily wear and charging
Typical failure mode
- Wi-Fi sensing
- Metal-dense rooms; several people moving at once
- Cameras
- Occlusion, low light, privacy-masked areas
- Wearables
- Forgotten, flat battery, taken off for a shower
Wet rooms such as bathrooms
- Wi-Fi sensing
- Coverable without fitting a lens
- Cameras
- Rarely fitted
- Wearables
- Usually removed
| Wi-Fi sensing | Cameras | Wearables | |
|---|---|---|---|
| What it measures | Disturbance of a signal in the room, turned into a posture estimate | Image | Acceleration and orientation of the device |
| Bedroom privacy | No camera in the room; only a posture estimate is produced | Requires images, usually with masking or blurring | No images involved |
| Resident cooperation | None required — nothing is worn | None to wear, but consent to being filmed | Daily wear and charging |
| Typical failure mode | Metal-dense rooms; several people moving at once | Occlusion, low light, privacy-masked areas | Forgotten, flat battery, taken off for a shower |
| Wet rooms such as bathrooms | Coverable without fitting a lens | Rarely fitted | Usually removed |
What "requires staff verification" means in practice
Every alert starts in a state that requires verification. When a member of staff arrives, they either resolve it or mark it a false alarm; both outcomes feed back into how sensitive that room is treated afterwards. The design does two things: it keeps responsibility with trained people, and it stops the system being switched off after the first false alarm.
In practice that means the workflow has to be written down: who receives the alert, how long they have to attend, what they record on arrival, and how a false alarm is marked. A system without that workflow is muted within a fortnight.
What a family should ask a care home
If you are looking at a care home, or the home has just installed monitoring, these questions say more about how it works in practice than any product sheet. They are about process, not specification.
Seven questions worth asking in person
- After an alert, who is responsible for attending? Which shift takes it overnight?
- What is the home's current target time from alert to someone at the bedside?
- If an alert is judged a false alarm, how is that recorded, and does the record change sensitivity?
- Which areas are covered by sensing and which are not? Does that include bathrooms and toilets?
- Does the system record video or audio? If so, how long is it kept and who can view it?
- Which records can a family ask to see, and in what form are they provided?
- If the home changes system later, what happens to the records already collected?
What the platform actually measures
SteadiCore SI produces a continuous posture estimate at room level and derives three signals from it: fall-like events, prolonged stillness, and leaving the bed without returning. Each signal carries a timestamp and a room location, which together make a timeline that can be reviewed afterwards.
At the home level, those events are aggregated by zone, time of day and shift, so a manager can see which bathroom or which night shift produces more alerts. These are operational measures, not clinical ones: they describe pressure points in staffing and environment, not the health of any individual resident.
Privacy and data handling
The design rules are: no camera in the room, no facial recognition, no image recorded. What is processed is a posture estimate plus event timestamps, and residents are referenced by pseudonym rather than by name as the primary identifier in the system.
In Hong Kong, the six Data Protection Principles in the Personal Data (Privacy) Ordinance (Cap. 486) apply to residents' personal data — purpose of collection, accuracy, restriction of use, security, openness, and access and correction. For a home using SI (artificial intelligence AI) tools on personal data, the Privacy Commissioner's "Artificial Intelligence: Model Personal Data Protection Framework", published in 2024, sets out the recommended practices for procuring, implementing and using them.
The Hong Kong regulatory context
Residential care homes for the elderly in Hong Kong are regulated under the Residential Care Homes (Elderly Persons) Ordinance (Cap. 459), with the Director of Social Welfare as the licensing authority; a home must hold a licence to operate. The Social Welfare Department issues the Code of Practice for Residential Care Homes for the Elderly under section 22 of that Ordinance, setting out the principles, procedures and standards for operation and care. The Residential Care Homes Legislation (Miscellaneous Amendments) Ordinance 2023 was gazetted on 16 June 2023 to strengthen the monitoring and quality of homes.
For a home, that means two things. Any new monitoring measure has to fit the existing care and record-keeping workflow rather than sit beside it, and the record of an event has to survive later scrutiny: who was alerted and when, how long attendance took, what the outcome was. Fall prevention is not a feature of one product; it is part of how a home meets its care standard.
Five steps to run a fall prevention programme
The steps are in the order they have to happen, and they apply to any sensing method, not only Wi-Fi.
- 1Define the gap you are closingHow long does it currently take, on average, from an event to somebody knowing? Answer that from existing staff records first, so there is a baseline to compare against.
- 2Cover high-risk zones, not the whole homeBathrooms, night hours and dementia care zones are usually where coverage earns its keep first. A smaller scope is a scope whose workflow can actually change.
- 3Write the response workflow firstWho receives the alert, how quickly they attend, what they record, and when an alert is marked a false alarm. This has to exist before installation, not after.
- 4Pilot it and count the false alarmsLog every alert during the pilot, including the false ones. The false-alarm rate is what decides whether the system is still switched on in a fortnight.
- 5Connect the record to review and complianceThe event timeline has to be exportable for internal review, family conversations and licensing inspection.
SteadiCore evidence note
If this was useful, the next note comes straight to your inbox
What arrives: one short note on what public research and Hong Kong care-home practice actually say about fall risk, plus a printable risk-assessment sheet you can walk a floor with — the same sheet for operators and for families.
Common questions
- Can a fall detection system guarantee that every fall is detected?
- No. Every sensing method has blind spots. The system raises a possible fall alert that staff have to verify; it is not a confirmation of a fall and it does not replace staff judgement.
- How does it know what is happening in a room without a camera?
- Wi-Fi sensing reads how the signal is disturbed in the space and derives a posture estimate (RF Pose Estimate). What comes out is posture and events, not an image. No facial recognition is used.
- Does the resident have to wear or operate anything?
- No. Wi-Fi sensing requires nothing worn and nothing charged daily, which is why it still works in the bathroom and at night.
- Will there be a lot of false alarms?
- It depends on the room and the sensitivity setting. Log every alert and its outcome during a pilot and tune sensitivity from real data rather than aiming for zero false alarms from day one.
- Can the records be used for licensing review or family conversations?
- The event timeline can be exported, including alert time, attendance time and outcome. It is part of the home's care record, not a medical record.
Sources
- Residential Care Homes (Elderly Persons) Ordinance (Cap. 459)(opens in a new tab)Licensing regime and operating requirements for care homes for the elderly.
- Social Welfare Department — Code of Practice for Residential Care Homes for the Elderly (June 2024 revised edition)(opens in a new tab)Issued under section 22 of Cap. 459; sets the standards for operation and care.
- Personal Data (Privacy) Ordinance (Cap. 486)(opens in a new tab)The six Data Protection Principles that apply to residents' personal data.
- PCPD — Artificial Intelligence: Model Personal Data Protection Framework (2024)(opens in a new tab)Recommended practice for procuring, implementing and using AI tools.