One false alarm every three months thanks to AI

What is a chair? For a human being, that is a dead simple question. Four legs or not, armrests or not, when you see one you know. For a computer, it took decades before programmers could finally create a program that could distinguish a chair from other objects with some accuracy.

It seems trivial. But Harro Stokman, who worked on exactly this issue as a doctoral student at the University of Amsterdam, discovered that triviality is an illusion. With a few colleagues, he founded Euvision, a company that made headlines with an app that could classify photos on your phone. Dogs, cats, sailboats. In 2014, the American company Qualcomm bought the company.

After a few years with the tech giant, it was time for something different. AI, meanwhile, had gotten a lot smarter. It was now possible to infer from moving images what people were doing. But Stokman sought an application that mattered socially.

Three nurses for one hundred residents

The eye fell on elder care. In Dutch nursing homes with 100 residents, often only three nurses work at night. These come into action when a sensor goes off. Often it turns out to be a false alarm. A pillow falling off the bed sends the same signal as a client falling. The protocol also dictates going into each room a few times a night, just to check that all is well.

Both scenarios are disruptive. For the resident because of the broken night. For the caregiver who is called up unnecessarily. And for society, because this way of working takes a lot of manpower that isn’t there.

The numbers don’t lie. The staff shortage in Dutch healthcare will quadruple in the next decade, from 66,400 in 2025 to 265,600 in 2034. [1] Nursing homes specifically are already short 14,200 people. [2] A quarter of current care staff will retire within five years, while the number of elderly people continues to grow.

“They did work with cameras and sensors in that sector, but it was all very old-fashioned and they gave a lot of false alarms.” Harro Stokman [3]

The sensor that understands what it sees

In 2018, Stokman founded Kepler Vision Technologies as a spin-off from the University of Amsterdam. The product is called Kepler Night Nurse. It is a smart sensor that hangs in the room and sounds an alarm when something is in danger of going wrong.

The difference with traditional sensors is not in what the system sees, but in what it understands. The AI recognizes whether someone is lying down, about to fall, walking around, sitting on the floor, or leaving the room. It sees the difference between someone going to the bathroom and someone who actually needs help.

And the false alarms? One per sensor every three months. [3]

That sounds like a technicality. But for a nurse who is disturbed dozens of times during a night shift by alarms that mean nothing, it is the difference between exhaustion and workable nights. Alarm fatigue is a real problem in healthcare. After hundreds of false alarms, you react more slowly, or not at all.

No camera, but reassurance

A sensor in your bedroom that’s always on. That sounds like a privacy nightmare. But Stokman turned the argument around.

The system does not recognize people, only activities. The sensor does not send images, only a text message when something is wrong. And most importantly, there is no longer a caregiver coming into the room every few hours to check that all is well.

For many residents, this actually improves their privacy. No more physical intrusion at night. No flashlight in the face. Supervision only when needed.

“The system does not recognize people, only activities. A timely alarm can prevent a lot of misery, but just as important is the peace of mind it brings to both resident and caregiver.” Harro Stokman [3]

From university to nursing home

Kepler now monitors more than 14,500 clients and operates in 26 European countries. [4] Turnover was about one million euros in 2023. Stokman plans to triple each year over the next three years. With a €1.5 million investment from ROM InWest and earlier funding from UvA Ventures and the Ministry of Economic Affairs, the company has now raised more than €9.8 million. [4][5]

Healthcare is often cautious about new technology, Stokman notes. But that there is now competition, he thinks, is a good sign. It means the market is taking the problem seriously.

The shift from a purely technical company to a healthcare partner required new faces who speak the language of the market. With the arrival of Stephanie van Rosmalen as CMO and Lex Erades as COO, Kepler has professionalized its team as they continue to grow. They must ensure that the technology not only works in the lab, but is actually embraced on the nursing home floor.

Where science touches the workplace

The story of Kepler Vision shows what is possible when academic expertise meets a concrete problem. Stokman began by asking how a computer recognizes a chair. Twenty years later, that same knowledge helps a nurse at 3 a.m. know when to walk and when not to walk.

For the Dutch AI scene, this is a lesson. The most valuable applications are not always in the latest models, but in the translation of existing knowledge to places where it really makes a difference. A spin-off from the UvA that helps nursing homes. An algorithm that once recognized chairs and now monitors human lives.

The complexity is not in the code. It’s in finding the right problem.

This article was published on Dutchstartup.ai.