Falls are one of the most serious risks for older people living alone, and the longer someone lies on the floor undiscovered, the worse the outcome. That is why "sense a fall without a camera and without a wearable" is one of the most requested Wi-Fi sensing ideas. The physics genuinely supports it. The reliability is where honesty is required.
01Why camera-free fall detection matters
The two existing options both have real drawbacks. A camera can detect a fall, but a camera in a bathroom or a bedroom is exactly where people least want one, and it records an image of everything else too. A wearable pendant with a button or an accelerometer works only if the person is actually wearing it and, for a button, is conscious and able to press it — often not the case after a bad fall.
A Wi-Fi link sidesteps both problems. It senses movement in the room from the radio channel, as explained in our complete CSI guide, so there is no image of the person and nothing to wear or charge or remember. For a space someone owns, that combination of no camera and no wearable is genuinely compelling, which is why the idea keeps coming up.
02What a fall looks like in CSI
A fall is not just "movement". It has a characteristic shape in the channel that distinguishes it from ordinary activity, and that shape has two parts.
- A large, fast transient. A body dropping to the floor moves a lot of mass quickly, disturbing many of the signal paths through the room at once. In the CSI energy this is a spike that is both bigger and steeper than walking, sitting, or gesturing.
- A period of unusual stillness. After a real fall the person typically stays down. So the spike is followed by a quiet floor — much less movement than normal activity, held for an unusually long time.
It is the combination that matters. A spike alone is just vigorous movement; stillness alone is just an empty or resting room. A sharp spike followed by prolonged stillness is the signature a fall detector hunts for. This is almost the opposite of the slow, steady micro-motion used in breathing detection — here the interesting event is the sudden large change, not the tiny regular one.
03A simple detector, and why it is not enough
You can build a first attempt with the same tools as a CSI motion alarm. Compute the short-term energy or variance of the CSI, as the alarm does, then add a small state machine: watch for a transient that crosses a high threshold, and then check whether the signal drops to and stays at an unusually low level for some seconds afterward. If both happen in sequence, flag a candidate fall.
This simple detector demonstrates the principle, and in a tidy, single-occupant room it can even work tolerably. But move it into a real home and it starts crying wolf, because a lot of harmless events produce the same spike-then-quiet shape.
04The false-positive problem
This is the heart of the difficulty, and it is worth being blunt about. Many everyday events look like a fall to a single CSI link:
| Event | Why it mimics a fall |
|---|---|
| Sitting down heavily | A fast, large disturbance then stillness in a chair |
| Dropping a bag or a box | A sudden spike with no person down at all |
| A door slamming | A brief large transient across the channel |
| A pet jumping down | Fast motion followed by the animal settling |
| Lying down to rest | Movement to the floor or bed, then prolonged stillness |
Every false alarm erodes trust, and a fall detector nobody believes is worse than none, because the real alert gets ignored. The opposite error is just as serious: a slow slide to the floor, or a fall partly shielded by furniture, may never produce a clean spike, so it is missed entirely. A single link, using a hand-set threshold, cannot reliably separate the real falls from the convincing imposters. That is the ceiling of the simple approach.
05Why real systems need training data
Because a threshold cannot capture the difference between a fall and a heavy sit, serious fall detection turns to machine learning. Instead of one number, the system extracts a set of features from the CSI around each candidate event — how fast the spike rose, how high, how the energy was spread across subcarriers, how deep and how long the stillness that followed — and a trained classifier decides "fall" or "not a fall".
That classifier is only as good as the data behind it, and this is the catch. It needs many examples of real falls and, just as importantly, many examples of the near-misses: the heavy sits, the dropped bags, the lying down. Collecting falls safely and realistically is hard, every room changes the signal, and a model trained in one layout often degrades in another. The data problem, not the physics, is what keeps Wi-Fi fall detection in the research and prototype stage rather than in a reliable consumer product. For the broader picture of turning CSI into labelled activities, see CSI activity recognition.
06Honest accuracy expectations
Published research prototypes often report high detection rates, but those numbers come from controlled rooms, scripted falls, and tuned setups. A real home has furniture, pets, other people, and layouts the model never saw, so the honest expectation is noticeably lower and more variable than any single headline figure. The useful framing is a helpful hint that can prompt a check-in, not an alarm you would bet a life on. Built that way — in your own space, as an experiment, as one layer among several — CSI fall sensing is a worthwhile project. Sold as a guaranteed safety net, it is a dangerous overclaim.
07FAQ
How does a fall look different from normal movement in CSI?
A fall is a large, fast burst of change in the channel, bigger and more sudden than walking, followed by a period of unusual stillness as the person stays down. It is that specific shape, a sharp spike and then a quiet floor, that a detector looks for rather than movement alone.
Why is Wi-Fi fall detection attractive for elder care?
Because it needs no camera and no wearable. A camera in a bathroom or bedroom is intrusive, and a pendant only helps if the person is wearing it and can press it. A Wi-Fi link in the room senses a fall without recording any image and without asking the person to carry anything.
What causes false alarms in CSI fall detection?
Anything that produces a big, sudden disturbance then goes quiet can imitate a fall: sitting down heavily, dropping a bag, a door slamming, or a pet jumping off furniture. Telling these apart from a real fall is the central difficulty, and it is why a simple spike detector is not enough on its own.
How accurate is Wi-Fi CSI fall detection really?
Honestly, it varies a lot with the room, the setup and the training data, and a single link is best treated as a helpful hint rather than a guarantee. Research prototypes report high numbers in controlled rooms, but real homes are messier, so expect both missed falls and false alarms and never rely on it alone for safety.