Camera fall detection in a care home: what it sees, and what it costs
A camera fall detection system compares video frames to judge that a person has gone to the floor. It works in a lit common room and fails behind a curtain or in the dark, and a bedroom camera raises consent and dignity questions, so most care homes keep cameras in shared areas and verify each alert by hand.
By SteadiCore
What a camera fall detection system actually measures
A camera does not see a fall. It sees a sequence of frames, and the model behind it has learned that certain sequences — a body shrinking down the frame, a sudden change from upright to horizontal, then very little movement — usually mean someone is on the floor. Everything the system reports is an inference drawn from pixels, which is why the honest output is a possible fall rather than a confirmed one.
That inference is only as good as the view. A camera mounted at ceiling height in a corridor sees a wide, clear scene: walking, turning, standing up from a chair. The same camera pointed into a bedroom sees a bed, a curtain, a chair back and a person who is often under a blanket. The model has less to work with, and it knows it less often.
Why camera fall detection struggles in a dark bedroom
Infrared illumination lets a camera keep working after lights out, but it changes what the model sees: faces become flat white shapes, dark clothing and dark floors merge, and a blanket on the floor looks much like a person. A night-time alert built on that image is a prompt to go and look, not a fact.
Occlusion is the harder problem. A privacy curtain around a bed is standard practice in Hong Kong care homes and it is exactly the object that removes the camera's line of sight. Residents also cover cameras, and staff sometimes do. A system whose coverage depends on a curtain staying open is a system with an unknown blind spot, and a blind spot nobody measures is worse than a documented one.
Camera fall detection, privacy and consent in a Hong Kong care home
Video of a resident is personal data, and continuous video of a bedroom is personal data of the most sensitive kind. The Personal Data (Privacy) Ordinance (Cap. 486) applies to a care home as it applies to any data user: collect only what you need, tell people what you are collecting and why, keep it no longer than necessary, and restrict who can watch it. The Privacy Commissioner for Personal Data publishes guidance on CCTV surveillance that a home should read before it buys anything.
Consent is harder than a signed form. A resident living with dementia may not be able to give meaningful consent to a camera above their bed, and a family member's signature is not the resident's consent. That is the practical reason many operators draw the line at corridors, entrances and shared lounges, where the purpose is safety of a shared space rather than surveillance of one person's private hours.
Camera fall detection compared with the alternatives
The decision is not camera or nothing. Each method has a different failure mode, and the useful question is which failure mode your building can absorb. The table below compares what each approach physically measures, not what a brochure says it achieves.
Sensing methods for fall-like events in an aged care setting
Ceiling or wall camera
- What it measures
- Video frames and estimated pose
- Works in a dark bedroom
- Infrared only, image degrades
- Worn by the resident
- No
- Bedroom privacy
- Continuous image; needs a written basis
- Fits best in
- Corridors, lounges, dining rooms
Wearable pendant or watch
- What it measures
- Acceleration and impact at the wrist or chest
- Works in a dark bedroom
- Yes
- Worn by the resident
- Yes, and it must be charged
- Bedroom privacy
- No image
- Fits best in
- Residents who reliably keep it on
Bed or chair pressure sensor
- What it measures
- Weight leaving or returning to a surface
- Works in a dark bedroom
- Yes
- Worn by the resident
- No
- Bedroom privacy
- No image, low detail
- Fits best in
- Unplanned bed exit, night rounds
Wi-Fi sensing nodes
- What it measures
- Disturbance of Wi-Fi signals between nodes
- Works in a dark bedroom
- Yes
- Worn by the resident
- No
- Bedroom privacy
- No image; posture estimate only
- Fits best in
- Bedrooms and bathrooms with no camera
| Approach | What it measures | Works in a dark bedroom | Worn by the resident | Bedroom privacy | Fits best in |
|---|---|---|---|---|---|
| Ceiling or wall camera | Video frames and estimated pose | Infrared only, image degrades | No | Continuous image; needs a written basis | Corridors, lounges, dining rooms |
| Wearable pendant or watch | Acceleration and impact at the wrist or chest | Yes | Yes, and it must be charged | No image | Residents who reliably keep it on |
| Bed or chair pressure sensor | Weight leaving or returning to a surface | Yes | No | No image, low detail | Unplanned bed exit, night rounds |
| Wi-Fi sensing nodes | Disturbance of Wi-Fi signals between nodes | Yes | No | No image; posture estimate only | Bedrooms and bathrooms with no camera |
Why a camera alert still needs staff verification
A camera will report a fall-like pattern when a resident lowers themselves carefully to the floor to reach something, when two people hug, when a mattress is being changed, or when a wheelchair is folded. These are not model defects; they are the visual signature of a body moving down, and the model cannot read intent. Every alert therefore ends with a person walking to the room and looking.
That is not a weakness to be engineered away. The value of any alert is the time between the event and a human arriving, and the record that the arrival happened. A home that measures that interval, and the share of alerts that turned out to be nothing, learns more about its night shift than any accuracy figure on a datasheet.
How to decide whether camera fall detection belongs in your building
Start with the room, not the product. List the places where a fall is both likely and unseen — a corridor at 03:00, a bathroom doorway, the space beside a bed — and ask which of them you are willing to put a lens in. Areas where a resident is dressed, washed or sleeping are usually the answer to that question, and they are also where many falls happen, which is the tension worth naming in your procurement notes.
If a camera is the right call for a shared area, write down four things before signing: the purpose and the rooms covered, who may view footage and for how long it is kept, how a resident or family member can object, and how the alert reaches staff who are already busy. A platform such as Elyvo can take camera or Wi-Fi input and route the resulting alert into one workflow, which matters more than which sensor produced it.
Whichever route you take, keep the language accurate. A camera infers a fall-like pattern; it will not catch every fall, and it never replaces the judgement of the person who opens the door. If you are still mapping methods, the comparison in our sensor guide goes one level deeper into the hardware, and the SteadiCore home page shows how the four sensing methods feed one response loop.
If the open question is what staff should do between alerts rather than how to sense them, the same group publishes caregiving coordination material at Easikin (opens in a new tab) and cognitive-activity material at Sanospark (opens in a new tab).
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.
Questions people ask
- Does camera fall detection work in the dark?
- Infrared illumination lets most cameras produce an image with the lights off, but that image is lower in detail and a blanket or dark clothing can hide a person on the floor. Treat a night alert as a prompt to check the room, not as confirmation.
- Can a care home put a fall detection camera in a resident's bedroom?
- It can be lawful, but it needs a documented purpose, a genuine basis for the resident's consent, signage, restricted access and a retention limit. Where a resident cannot consent, many homes choose not to install one and use a sensing method that produces no image instead.
- Is camera fall detection more accurate than Wi-Fi sensing?
- There is no measured head-to-head figure we can quote, and the two fail differently: a camera loses the room behind a curtain, Wi-Fi sensing loses precision but keeps working through fabric and in the dark. Choose on the failure mode you can staff around.
- What should happen when a camera raises an alert?
- A staff member goes to the room and verifies. The alert should carry the room, the time and the confidence level, and the visit should be recorded so the home can show what it did and when.
Sources
- Office of the Privacy Commissioner for Personal Data — CCTV surveillance guidance (opens in a new tab)The regulator's guidance on collecting images of individuals, including in workplaces and care settings.
- Personal Data (Privacy) Ordinance (Cap. 486), Hong Kong e-Legislation (opens in a new tab)The statutory data protection principles a care home must apply to video of residents.