CCTV Camera Motion Detection: How to Stop False Alarms
CCTV camera motion detection works in one of three ways: the camera or recorder compares video frames and fires when enough pixels change, a passive infrared (PIR) sensor notices moving body heat, or AI checks whether the thing moving is a person or a vehicle (sold as person detection). The first is on almost every system and causes most false alerts, because rain, insects and headlights change pixels too.
This guide explains each kind, why the alerts flood in, what to change in the settings, what to look for when buying and when person detection is the fix. We build AI for existing CCTV, and plain motion detection is still the first check in the weapon detection we built for more than 1,000 cameras. It is just the wrong thing to send to your phone.
The three kinds of motion detection
| Kind | Reach | What fools it |
|---|---|---|
| Pixel change: any change in the picture | No fixed range: the object has to change enough pixels | Rain, insects, headlights, shadows, trees, the switch to night mode |
| PIR sensor: heat moving across its view | About 5 to 12 m (16 to 40 ft) on cameras | Sun, warm air, hedges cooling at dusk; blind through glass |
| Person detection (AI): the shape of a person or vehicle | About 10 to 30 m (33 to 100 ft), depending on the lens and model | Distance, bad angles, anything person-shaped |
Makers blur the names. "Smart motion detection", "human detection", "humanoid detection" and Hikvision's "Motion Detection 2.0" are all person detection. "Advanced motion detection" can be either pixel change with extra filters or person detection, so check what it says it ignores.
How pixel motion detection works
The camera or the recorder compares each frame of video with the one before. If enough of the picture has changed, it calls that motion. It then does whatever you set: start recording, mark the timeline, send a notification. On older DVR systems the recorder does the comparing; most IP cameras can do it themselves.
Two settings decide what counts, and most menus hide both behind one "sensitivity" slider:
- How much a pixel must change. In Frigate, an open-source recorder that documents its workings, a pixel counts as changed by default when its brightness moves by 30, on a scale of 1 to 255. Set that lower and, in Frigate's words, it starts to pick up "grass blowing in the wind, shadows".
- How many changed pixels in one clump. Changed pixels are grouped, and a group below a minimum size is ignored. Set that higher and small animals stop triggering, but so does a person far from the camera.
Nothing in this process knows what moved. That is the whole problem, and the figure below shows why no sensitivity setting gets you out of it.
Why motion detection gives false alarms
We worked out how much of the picture different things cover on a common 4MP camera with a 2.8mm lens. A person 20 m away fills 0.09% of the picture. A moth 10 cm from the lens fills 3.2%, more than 35 times as much, and at night it is lit by the camera's own infrared lamps. Even a person 5 m away, at 1.4%, covers less of the picture than the moth.
So turning sensitivity down removes the person at the fence before it removes the moth. The things you want to catch are often smaller in the picture than the things you do not.
The same sums explain missed people. Axis notes in its motion detection manual that "small and distant objects might not be detected" and that someone walking straight towards the camera has to cover more ground before triggering than someone crossing the view. Contrast matters too: dark clothes against a dark yard change each pixel very little, so the change can stay under the threshold.
The usual causes, with who documents them:
- Insects and spider webs at night. Swann's support note says insects are drawn to the infrared LEDs cameras use to see in the dark, and webs collect around the lens.
- Rain, snow and wind. Axis warns in its motion detection manual that heavy rain or snow affects accuracy and that a camera shaking on its mount "might trigger false alarms". Swaying trees and bushes do the same.
- Light. Headlights sweeping a wall, fast shadows, clouds, reflections in windows and wet tarmac.
- The switch between day and night mode. The whole picture changes when the infrared comes on at dusk. Frigate has a setting just to ignore a frame when most of it changes at once.
- A busy view. A public road or pavement in the picture means an alert for every car and passer-by.
Indoors with steady lighting, plain motion detection is fine. Outdoors at night it is at its worst, which is exactly when most businesses want alerts.
What is PIR motion detection?
PIR stands for passive infrared. It is a separate heat sensor, usually behind a small white plastic window, and it does not look at the video at all. It fires when something warmer than the background moves from one of its zones to the next. Battery cameras use it because it sips power: the camera sleeps until the PIR wakes it.
It ignores leaves, rain and headlights, which is why it gives fewer false alarms than pixel change. Its limits:
- Short reach. Reolink lists up to 5, 8 or 10 m for most of its battery cameras, depending on the sensitivity level. INSTAR quotes about 5 m for its indoor cameras and 12 m outdoors.
- Blind through glass. INSTAR's installation notes say motion detectors "do not detect movement behind glass", so a PIR camera on a windowsill misses the car park outside.
- Weak head-on. It detects someone walking across its view sooner than someone walking straight at it. Mount it side-on to the path people take, 1.8 to 2.5 m up by INSTAR's figures.
- Its own false alarms. Direct sun, warm air from a vent, and hedges or walls that release the day's heat at dusk.
PIR is not the same as the camera's IR. IR on a spec sheet usually means the infrared lamps for night vision. They light the scene; they detect nothing. Some cameras let you require both PIR and pixel change before alerting. That cuts false alerts, at a price: anything only one of the two notices, such as a person beyond the PIR's reach, no longer alerts.
How to stop motion detection false alarms, in order
- Count first. Note the alerts per camera per night for a week. Often a few cameras cause most of them.
- Fix the view. Brush off webs, tighten the mount, and re-aim so trees close to the lens and the public road are out of the picture. For person detection Hikvision recommends mounting at 3 to 5 m, tilted down about 10°.
- Draw zones. Detect only inside the area you protect, and mask out what moves on its own, such as trees and flags. Axis advises keeping the zone clear of a nearby road or pavement, because an object only partly inside the zone still triggers. Use privacy masks to black out a neighbour's garden or windows altogether; our GDPR guide covers cameras that see the street or a neighbour.
- Set sensitivity with a walk test. Walk across the farthest spot you care about, at night, and lower sensitivity only until that walk stops triggering. Then go back up one step.
- Use time and size filters if you have them. Axis's motion detection app can ignore objects that appear only briefly ("light beams from a passing car"), small objects and objects that sway without going anywhere.
- Use separate day and night settings where the camera allows profiles. One setting rarely suits both.
- Schedule alerts for when the site should be empty.
- Send alerts from person detection, not motion. Keep motion for marking the recording. This is the step that changes the numbers; the rest only trims them.
If steps 1 to 7 leave you with more than a couple of alerts per camera per night, stop tuning. You have reached the limit of what pixel counting can do.
Person or humanoid detection: what it fixes, what it does not
Person detection, which budget cameras often label humanoid detection, adds one step: after something moves, a trained model looks at it and decides whether it is a person, a vehicle or neither. Only the first two alert. A moth that fills 3% of the picture is usually ignored, because it does not look like a person.
| What the menu calls it | Where you see it |
|---|---|
| Motion Detection 2.0, AcuSense | Hikvision cameras and recorders |
| SMD (Smart Motion Detection), WizSense | Dahua cameras and recorders |
| Human detection, person detection | Amcrest, Reolink, most consumer cameras |
| Humanoid detection, humanoid tracking | Budget Wi-Fi cameras and their apps, Tenda among them |
Dahua describes its current SMD as filtering out "small and large animals". Our guide to Hikvision AcuSense covers how to tell whether kit you already own has it.
You can add it four ways without replacing every camera:
- Swap the worst one or two cameras for models with person detection built in.
- Swap the recorder for one that analyses older cameras' video. Check its channel limits: some run the smarter rules on only one or two cameras.
- Add a box or cloud service that reads the video streams your cameras already send.
- Have custom detection built when you need something the products do not offer.
Our AI CCTV guide compares those four routes and their costs.
Person detection has limits of its own:
- Range. It needs enough pixels on the person. Hikvision gives 10 to 22 m for its first AcuSense cameras and 20 to 30 m for the generation launched in 2020, depending on the lens and model.
- It still gets things wrong. Hikvision's own guidance lists headlights, reflections and objects close to the camera as remaining causes of false alarms.
- It cannot tell a thief from the night cleaner. A person is a person. You still need zones and a schedule, and rules such as line crossing (an alert only when someone crosses a line you draw) help more than a wide "any person" alert.
It does not change your legal position either. Under UK GDPR the footage is personal data with or without it; our CCTV and GDPR guide covers the duties.
What to look for when buying
Ignore "motion detection" on the box: every camera has it. Look for person and vehicle detection that runs on the camera itself, zones you can draw, a schedule, and alerts you can set to people only. For a battery or wireless camera, check the PIR range against the distance you need to cover. For anything outdoors, check how far the person detection reaches at night, not just the night vision range.
Our AI security camera buying guide compares the options with prices, and how far to trust each CCTV analytic rates what comes after person detection, such as line crossing and loitering.
Should cameras record only on motion?
No. Motion-only recording saves disk space, but it fails in the worst way: the camera fails to record a distant figure for the same reason it fails to alert on one, and you find out after the theft. It also drops the minutes before an event, which often show how someone got in.
Record around the clock and use motion and person events as bookmarks to jump to. The cost is storage. At a constant 3 Mbps, the high-quality rate Verkada lists for its cameras up to 5MP, one camera writes about 32 GB a day, so 16 cameras kept for 30 days need about 15.5 TB. A lower quality setting and H.265, a newer way of compressing video that most recorders support, bring that down. If you must record on motion, set a pre-record buffer of several seconds so the clip includes the approach. Our cloud CCTV guide works through the storage and upload numbers.
If alerts go to a monitoring centre or the police
False alerts stop being an annoyance and start costing money once someone is paid to look at each one, and in the UK they can cost you police response. Monitored CCTV that calls the police needs a unique reference number (URN) from the local force, and the system must meet BS 8418, the British Standard for detector-activated CCTV.
The National Police Chiefs' Council requirements (updated June 2026, for forces in England, Wales and Northern Ireland) set the rule: after three false calls in a rolling 12 months, police response is withdrawn. Getting it back means showing the cause was found and fixed and, under the conditions written for intruder alarms that CCTV follows, 90 days free of false calls.
One line catches people out. A request for police to attend "sightings of individual(s) seen on protected premises where no criminal activity is in progress" is treated as civil trespass and counted as a false call. So a monitored system needs detection that operators can trust and a clear view of what the person is doing, not just proof that something moved.
Is plain motion detection still worth using?
Motion detection is a poor alarm and a good filter. It is cheap to run, so it is the right first check before anything expensive. Frigate works this way by design: its documentation calls motion "a first line check to see if there is anything happening in the frame worth checking with object detection". It is level one of the four levels of AI in security cameras, and the higher levels still stand on it.
We built weapon detection for Angel Protection, a US school and hospital security firm, on the same idea. Cheap processors decode the video and run a plain motion check on every camera, and only frames with movement reach the graphics card running the weapon model. That design cut hardware cost by 90%, and the system runs on more than 1,000 CCTV cameras with about a second from event to alert. The weapon detection case study has the detail.
For a UK CCTV security company we built real-time trespass and violence detection that reads the streams of the cameras already on site, with 15 to 20 cameras per RTX 5060 Ti graphics card and the footage staying in the building. That is the route to take when "a person is present" is not a useful alert, for example a yard where staff work nights and only someone climbing the fence matters.
If built-in person detection solves your problem, use it and keep your money. If it does not, a feasibility study on your own footage (£4,000 to £6,000, credited to the build) tells you whether custom detection will work before you commit.
Questions
Do CCTV cameras have motion detection?
Nearly all do. Pixel-change motion detection is built into almost every recorder and IP camera, including old analogue systems. Whether it is switched on, and whether it only marks the recording or also sends alerts, is a setting.
Do motion cameras record all the time?
It depends on the type. Battery cameras sleep and record only when triggered. Wired CCTV can record continuously, on motion only, or continuously with motion marked on the timeline. We recommend the last.
How far can a motion detector detect?
A PIR sensor on a camera reaches roughly 5 to 12 m (16 to 40 ft). Pixel motion detection has no fixed range: it depends on how many pixels the object changes, so a person 20 m away on a wide lens may not trigger. AI person detection works to roughly 10 to 30 m (33 to 100 ft) depending on lens and model.
Which is better, a PIR sensor or an IR sensor?
They do different jobs. A PIR sensor detects moving heat and triggers the camera. IR on a camera normally means infrared lamps that light the scene at night. Many cameras have both.
What is humanoid detection in CCTV?
It is another name for person detection, common in budget camera apps: the camera alerts only when it sees a human shape. "Humanoid tracking" is different. Tenda describes it as the camera turning to keep the detected person in the centre of the frame.
What is motion tracking in CCTV?
A camera that can pan and tilt turns by itself to follow whatever is moving. It is useful on a single camera covering a wide yard, but while it follows one person it is not watching the rest.
Why does my camera send motion alerts at night when nothing is there?
Usually insects or a spider web close to the lens, lit by the camera's infrared, or rain and headlights. Clean the lens housing, draw a tighter zone, and if the camera has person detection, send alerts from that only.
How do people trick a motion sensor?
Mostly they do not need to. Plain motion detection already misses someone far from the camera, in dark clothes against a dark background, or walking straight towards the lens. A PIR sensor misses anyone behind glass or beyond its reach. That is why we recommend alerting on person detection inside a drawn zone and recording around the clock.
Why did my camera record motion but not send an alert?
Detecting and alerting are separate steps. The camera may have seen the movement and then dropped it because it fell outside your zone or schedule, or a person-only filter decided it was not a person. If the event is in the log but never reached your phone, look at the app's notification settings and the site's internet connection, not the detection.
Does motion detection work through a window?
Pixel motion detection does, though the camera's infrared reflects off the glass at night and blinds it. PIR does not, because glass blocks the body heat it relies on. Mount the camera outside.