Dr. Harro Stokman, CEO and founder of Kepler Vision Technologies · Published 12 Apr 2026 · Updated 5 Aug 2026
Four kinds of sensor are used in fall prevention for the elderly, and this article starts by naming them. Then it goes a level deeper than most articles on the subject, to the part that decides whether a deployment actually works: what physically goes into the room, what it needs from your building, and what it takes to keep running.
The four kinds of sensor in use
Accelerometers, worn on the wrist or as a pendant. They measure sudden changes in motion and look for the signature of an impact. They work when worn, and when the fall is hard enough to register.
Camera sensors, mounted on the ceiling. We call ours smart sensors. Software interprets what the sensor sees — posture, movement, whether someone is on the floor. Nothing is required of the resident.
Pressure sensors, under the mattress or beside the bed. They answer one question reliably: is this person in bed.
Environmental sensors — passive infrared, door contacts, sound, vibration. They register that something happened in the room without establishing what.
Which of these suits your residents is a separate question, and I have written it out at length in the companion article. The rest of this page is about the camera option specifically: what it takes to put it in a working care home and keep it running.
What actually goes in the room
One smart sensor per bedroom, mounted on the ceiling, looking down. The lens is a fisheye, which matters more than it sounds: it sees the whole room rather than a cone of it, so there is no direction a resident can walk in and leave the picture.
Ceiling height does the rest of the work. Anywhere between about two and three and a half metres above the floor is right, which covers almost every care home room as built. Above that, the smart sensor should be hung lower on a pendant mount rather than left on the ceiling.
Two placement rules are worth knowing about, because they are the ones residents notice.
The smart sensor is not mounted directly above the bed. At night it illuminates the room with infrared, and the emitters give off a faint reddish glow. Lying in bed looking straight up into it is sometimes felt as unpleasant, so the smart sensor goes near the bed rather than over it.
The smart sensor is also kept away from the walls — roughly three quarters of a metre clear. Too close and the infrared bounces back off the wall, which leaves the far corner of the room darker than it should be.
Sunlight, artificial light and mirrors all need thinking about during installation rather than afterwards. A smart sensor aimed into a sun-facing window sees a white patch instead of a room. A mirror that reflects the bed can make the software believe there are two beds. None of it is difficult, but all of it is much cheaper to get right during installation than to come back and fix.
The bathroom is the question care organizations ask first, so let me answer it directly: there is no camera in the bathroom. What the software knows is that nobody is visible in the room and that the last person it saw was near the bathroom door. From that it reports her as being in the bathroom rather than as having left the room altogether. The one consequence for installation is that the smart sensor has to be able to see the bathroom door, which is one of the things that decides where in the ceiling it goes.
Night is a hardware question, not only a software one
Unwitnessed falls cluster at night, which makes night-time performance the specification that matters most. It is also the one buyers ask about least, probably because every supplier says their system works in the dark.
Working in the dark is a property of the sensor before it is a property of the software. A resident’s room at three in the morning may have no light in it at all, so the smart sensor provides its own, in the near-infrared: colour image during the day, infrared image at night. The illumination is built in. Nothing separate has to be specified, but the infrared does have a range, and that is the real reason the smart sensor is placed within a certain distance of the bed. Too far away and there is not enough infrared falling on the bed to tell whether somebody is in it.
One thing worth knowing. If a room is lit brightly enough at night, the smart sensor may not switch into night mode at all — and if there is that much light, there is no reason for it to. Infrared is there to replace light that isn’t present.
The practical consequence: when you evaluate systems, test at night in a dark room, not in a lit demonstration space during the day. If a supplier’s demo is daytime only, that tells you something.
Where the computation happens
This is the distinction that separates our two products, and it has consequences for what you buy.
Our server application runs on-premise. Video from the cameras reaches a server inside your own network, the neural networks run there, and alarms go out from there. One server covers a large number of rooms — up to around fifty cameras. It also works with cameras that are already installed, which makes it the practical route for an existing building.
Our embedded application runs on the sensor itself. The neural network sits on a chip a few centimetres behind the lens, and no video leaves the camera at all. Each unit is self-contained, which makes it straightforward to add rooms one at a time.
Sensor quality is not a detail
Our own published figures are dragged down by installations running on poor-quality sensors.[1]
The software can only work with what the sensor gives it. A cheap camera with a noisy image, aggressive video compression, or poor low-light performance produces worse detection than a good one running identical software. When a system underperforms, the sensor is at least as likely to be the cause as the algorithm.
We keep a list of camera models we have qualified, and we test new ones on request.
What the building and the network need
The software side of a deployment is fast. We update it centrally — thirty times over twenty months in the most recent period I have published figures for.[1]
The hardware deployment often takes more time, and it is where projects’ timelines may actually slip.
What tends to be needed: network cabling to each room, with the smart sensors drawing their power over that same cable rather than needing sockets; a firewall; and physical access to occupied bedrooms for mounting. That last one is the real constraint in a working care home. Rooms are occupied, residents have routines, and an installer cannot simply work through a corridor at will.
Your IT department will want figures rather than reassurances — bandwidth per camera, network segregation, which ports have to be open. We give those to your system integrator in writing, and go through them during technical training, so nobody is guessing on site.
If new smart sensors have to be mounted and cabled, the timeline is set by the electrician and by when you can get into residents’ rooms.
Where the alarm goes
A detection that does not reach the right person on the device she already carries is worthless. In practice that means integration with your nurse call system rather than another app for staff to check.
Our software is integrated with the nurse call and reporting platforms of our integration partners, among them Ascom, IQ Messenger, Televic, Indigocare, Ibernex, Phoniro, 9Solutions, Sensara, Skyresponse and Person Centred Software.[3]
If the system you run is not among them, we build the connection within 48 hours.
Ask any supplier for their integration list before you ask anything else about features. It is the cheapest way to shorten a shortlist.
Privacy
Camera sensors raise the sharpest privacy questions in this field, and they deserve a fuller answer than a paragraph, which is in the companion article. Three things belong here, because they are properties of the equipment rather than of policy.
We blur faces on your own premises, before any image leaves the building.
In the embedded version, images are processed on the camera and wiped immediately. The only thing retained is a log of the alarms sent.
And one point that is genuinely your decision rather than ours: some nurse call systems allow a caregiver to open the camera briefly after an alarm, to see the situation before walking in. Whether that is possible depends on the nurse call system you run, and whether it is switched on is for your organization to decide. It is useful — knowing whether a resident on the floor will need a patient lift (hoist) saves a trip — but it should be decided openly and told to residents and families, not discovered by them.
Deciding
Ninety-one care locations in the Netherlands run our software, plus sites in Belgium, Norway and beyond.[2]
If you are specifying a system, three questions will tell you more than a brochure: which cameras, tested in the dark, and integrated with which nurse call systems. You can run our embedded software on a Mobotix C71 free for a month, which is a better test than any answer I can give you here.
Contact us for a demo.
Frequently Asked Questions
How long does implementation of fall prevention sensors in a care facility take?
Implementation of fall prevention sensors takes an average of 2-4 weeks, depending on the size of the facility and the chosen system. Plug-and-play solutions can be operational within days, while more extensive systems require more time for installation, calibration, and staff training.
What are the costs of fall prevention technology and how do they justify themselves?
Investment in fall prevention sensors ranges from €200-800 per room, depending on the technology. These costs are quickly recouped through reduced fall incidents, lower insurance claims, and more efficient staff deployment. On average, each prevented fall saves €15,000-25,000 in medical costs.
Can residents with dementia or cognitive impairments cope with fall prevention sensors?
Camera-based fall prevention systems are ideal for residents with dementia because they require no active cooperation. The system works completely in the background without residents needing to consciously interact with it. Wearable sensors can be more challenging for forgetful residents.
How is care staff trained in using fall prevention systems?
Training of care staff consists of a practical session of 2-4 hours where they learn how to interpret notifications, operate the system, and follow protocols during alarms. Most suppliers offer ongoing support and refresher courses to ensure staff use the system optimally.
What happens if the fall prevention sensor gives a false alarm?
With a false alarm, staff is alerted but can quickly determine that there is no emergency situation. Modern systems learn from false alarms and adjust their algorithms to improve accuracy. It is important to register false alarms so the system can adapt to individual movement patterns.
Can fall prevention sensors also detect other emergency situations?
Yes, advanced fall prevention systems can often also detect other situations such as prolonged inactivity, unusual movement patterns indicating medical distress, or leaving the room at night. This multifunctional monitoring increases overall safety and provides broader protection for residents.
How do fall prevention sensors integrate with existing care systems?
Most modern fall prevention systems can integrate with existing nurse call systems, electronic patient records, and care management software via standard APIs. This integration ensures seamless workflows where alarms are automatically forwarded to appropriate staff and incidents are recorded in patient data.
Related Articles
- What is the difference between workload and work-related stress?
- What is the reliability of modern fall prevention systems for the elderly?
- What role does environmental modification play in fall prevention for older adults?
- How can I claim fall prevention physiotherapy?
- How does AI improve collaboration between healthcare and technology?
This content was generated with the help of AI and it may contain mistakes