Thermal sensors for home care: what they see, and what they miss

How a thermal sensor works

Everything warm gives off long-wave infrared light, between about 8 and 14 micrometres, and people give off a lot of it. A thermal sensor reads that light and turns it into a grid of temperatures (Teledyne FLIR). It sends nothing out, so it is passive, and it needs no lamp or daylight: it can be used in complete darkness (Infiniti Optics).

Two kinds are used in senior care. Thermopile arrays are small and cheap, from a single pixel up to about 80 by 80, and need no shutter. Microbolometer cameras typically start at 160 by 120 pixels, show far more detail and recalibrate themselves now and then with a small shutter (Calumino's technical brief).

What a person looks like at each resolution

Pixel sizes are our arithmetic from each sensor's published field of view, at the centre of the image, for an adult 45 cm wide and 1.7 m tall and a face 15 cm wide.

At 8 by 8 a person is a few warm pixels: enough to tell that someone is there, not what they are doing. A review of sensors up to 16 by 16 found that above that size a person's silhouette becomes clearly identifiable (a 2022 review). Resolution matters for falls too: in our own study, an AI reviewer caught 69% of staged falls at 32 by 24 pixels, 48% at 16 by 12 and 34% at 8 by 6 (How many pixels does a thermal sensor need to see a fall?).

Heat, not faces: what that means for privacy

A thermal sensor in a room shows no face. At the resolutions in the table a face spans a few pixels at most, while research on recognising faces in thermal images has worked from images of about 64 by 64 pixels of the face alone (Farag and colleagues, 2013). It does not show what anyone is wearing, and the room's objects appear only as warmer or cooler shapes.

Low resolution is not the same as anonymous, though. In a 2025 study under controlled conditions, a model identified people from 16 by 12 pixel thermal images about 95% of the time (a 2025 study). Heat images of a person are still personal data, and the client, or their representative, should agree to them.

We found no study comparing how older adults feel about thermal sensors and cameras. The nearest: in a survey of 304 caregivers of people with dementia, radar and light sensors were rated most acceptable and cameras and microphones least acceptable (Frontiers in Digital Health, 2023). Senecta's pods have a microphone, so the consent conversation should cover what it listens for and what is kept, which is nothing unless a fall is suspected. Our security and privacy page sets it out.

What a thermal sensor cannot see

• Through glass. Glass blocks long-wave infrared: point a thermal camera at a window and you see nothing on the other side (Teledyne FLIR). Acrylic and steam block it too (Panasonic), so a sensor cannot see into a glass or acrylic shower enclosure, and it sees only the room it is in.
• Behind furniture. A bed, a sofa or a person curled up on the floor can hide part of the body (a 2025 study). Where a sensor sits in the room matters as much as which sensor it is; one study found a ceiling sensor and a wall sensor together gave the best coverage (IEEE Sensors Journal, 2025).
• Other heat. Portable heaters can cause false detections, and one vendor's system is optimised for single-resident rooms without pets (Butlr's 2024 brochure); sensitivity drops when the room is close to body temperature (a 2025 conference paper).
• Several people. Methods struggle when more people are in the room than they were trained on (Sensors, 2025), and public thermal fall datasets show one person at a time.

Senecta's own limits are on How Senecta detects a possible fall.

What the evidence says about falls

The research is promising and young. A 2022 review of 15 studies of low-resolution thermal fall detection found most reached 85 to 90% or more on accuracy, precision, sensitivity or specificity, but every one was tested in controlled conditions, mostly empty rooms, with 1 to 10 participants (a 2022 review). DeepFall, from the University of Toronto, learned to flag falls as unusual events in thermal video (Nogas, Khan and Mihailidis), and later work on the same data improved on it (Mehta and colleagues, 2021). The public eHomeSeniors dataset was recorded with performing artists falling under a physiotherapist's guidance, and its authors note that laboratory fall datasets are performed by young people (Riquelme and colleagues, 2019).

Our own study tested an AI reviewer that was never trained on falls. It caught 69% of the staged falls of people it had never seen in that public dataset and called none of 499 non-fall clips a fall; over 76 hours in a lived-in room it made no false fall calls (95% upper bound, 1.2 a day) and caught 4 of the 9 staged falls the camera could see. The falls were staged, by younger adults, in one laboratory and one room, and it is our own study, not an independent evaluation (the full paper).

So ask any vendor, us included, for evidence from real homes, not only laboratories, and expect that no system catches every fall.

Thermal and the other camera-free options

Combining two weak signals is one answer to the weaknesses of each. Senecta pairs heat with sound, so neither has to decide alone: it looks for a warm shape that drops to the floor and stays there, with an impact or a cry at the same moment, and the pod asks the client before anyone is called. Fall detection options for home care agencies compares the categories in more depth.

Who uses thermal sensing in senior care

• Butlr sells ceiling thermal sensors to senior living and skilled nursing, more than 20,000 of them by early 2025 (MIT News), and has filed a trademark, Albie, for a home sensor we found no launch of yet. See Senecta vs Butlr.
• Calumino, in Australia, makes thermal sensors for fall activity monitoring with nurse call integration for care providers (Calumino).
• Senecta pairs a low-resolution thermal sensor with a microphone in plug-in pods with their own cellular link, sold to home care agencies for clients in their own homes. When it notices a possible fall, the pod asks the client “Are you okay?” and, if the answer is not a clear yes, calls your agency and the family. See how it works.

Other products avoid cameras with different sensors: all of them compared.

Questions agency owners ask

Can a thermal sensor see through walls?

No. Walls, and even ordinary glass, block the long-wave infrared a thermal sensor reads, so it sees only the room it is in, and not through a window or a glass shower screen. Radar is different: some radar sensors sense through walls.

Is a thermal sensor a camera?

It is a kind of camera that sees heat instead of light. At the resolutions used in senior care a person appears as a warm shape and a face spans a few pixels at most, so nobody sees a face; heat images are still personal data, and the client should agree to them.

How accurate is thermal fall detection?

Laboratory studies report high accuracy, but a review of 15 studies found every one was run in controlled rooms with 1 to 10 people. In our own study an AI reviewer caught 69% of staged falls in a public dataset and made no false fall calls over 76 hours in a lived-in room. No system detects every fall.

Do thermal sensors work in the dark?

Yes. They read the heat people give off and need no light, so they work the same at 3 a.m. as at noon.