How motion detection cameras work, from PIR sensors to pixel analysis, why they flood control rooms with false alarms, and how AI decides what actually moved.
Motion detection tells you something changed. It cannot tell you what.
A PIR sensor reads body heat, a camera compares pixels, and AI recognises people, vehicles and animals. They do different jobs.
Most motion alarms come from nature and light: trees, rain, insects under infrared, headlights and shadows.
Zones, size limits and sensible sensitivity cut the noise. Good placement cuts more.
Motion alone is rarely the right alarm trigger for a staffed control room.
The strong design: motion as a cheap trigger, AI to decide what moved, rules to decide what happens next.
On cameras that matter, record continuously and use motion to mark and search.
Search for motion detection and most results sell something for the home: motion sensor lights, alarm kits and doorbell cameras. They explain how a sensor notices movement, and stop there.
That skips the question a security head actually faces. A site with fifty or five hundred cameras does not lack movement. It has far too much. Every tree, car, cat and passing shower is movement. The real job is deciding which movement deserves a person's attention, and doing it before operators stop trusting the alarms.
This guide covers both halves: how motion detection works, in plain words, and how to build a system where motion starts the conversation and AI finishes it.
Motion detection is the automatic sensing of movement in an area, by a sensor or by software that analyses a camera's picture. A motion detection camera compares each new frame with the ones before it and raises an event when enough of the picture changes. Motion detection reports that something moved, not what moved.
That last point matters more than any specification. Wikipedia's entry on passive infrared sensors makes the same point about the most common sensor type: it detects general movement, but gives no information on who or what moved. Basic video motion detection has exactly the same limit. Almost everything in this guide comes back to that gap.
Buyers meet four distinct technologies under the same name.
Passive infrared (PIR) sensors. A PIR sensor watches for changes in infrared energy, which is radiant heat. When a warm body crosses from one sensing zone to the next, the signal jumps and the sensor trips. PIR sensors are cheap, draw very little power and are the standard detector in burglar alarms and motion sensor lights. Typical models cover about 10 metres. They see no picture and record nothing. Wikipedia's overview of motion detector technologies also covers dual technology units, which pair PIR with microwave so both must trip together.
Video motion detection in the camera. Most IP cameras can analyse their own picture and send an event when motion trips. It costs the recorder nothing. The catch is that every brand does it differently, with different settings, zones and quality, so on a mixed estate you manage each camera separately. ONVIF, the industry standard that lets cameras and software from different makers work together, gives these events a common shape: its Profile T covers motion alarm and tampering events, so one video management system can receive them from many brands.
Video motion detection in the video management system. Here the video management system (VMS) analyses each stream itself. Settings, zones and behaviour are the same on every camera regardless of brand, and the results feed straight into the same rules and alarms as everything else. It uses some server capacity per stream, which lightweight analysis keeps small.
AI object detection. AI does not look for change at all. It looks for things: a person, a car, a truck, a dog. It recognises a person standing perfectly still and ignores a branch thrashing in the wind. It needs more computing than motion detection, and like everything else it depends on the picture, but it answers the question motion cannot: what is it?
Video motion detection, often shortened to VMD, rests on a simple idea. The software compares the current frame with the frames before it. Where a pixel has changed brightness by more than a set amount, it is marked as changed. When enough changed pixels sit together, the software reports motion.
Four settings shape the result.
Sensitivity. How large a change in brightness counts. High sensitivity catches a dark figure against a dark background. It also catches sensor noise in a grainy night picture.
Size. How large the changed area must be. Set the minimum so that a person at the far edge of the area still qualifies, while raindrops and noise do not. Remember that an insect crawling on the lens can look enormous, which is why placement matters as much as settings.
Zones. Which parts of the picture are watched and which are ignored. Masking out a public road, a flag or a tree removes a whole category of alarms at a stroke.
Time. How long the change must last, and how long the system waits before it can fire again, called a cooldown. A sensible cooldown turns one event into one alert instead of a burst.
Some systems in the industry go further. They keep a model of the empty scene that slowly adapts to lighting and to repetitive movement such as leaves and waves, a family of techniques computer vision calls background subtraction. Even these struggle outdoors: Wikipedia's summary of foreground detection notes that wind, rain and changing light make outdoor scenes hard for methods that assume a static background.
Because pixels change for many reasons other than intruders, and basic motion detection cannot tell the difference. The usual causes are much the same on every site.
Trees, bushes and flags. Wind moves them all day, and a gust moves a large area of the picture at once.
Rain, snow and fog. Drops close to the lens look large and bright, especially under infrared at night.
Insects and spider webs at night. The infrared lights built into most cameras sit right beside the lens and attract insects. Spiders follow the insects and spin webs across the housing. A web glowing in infrared and swaying in a breeze is a classic cause of night time motion alarms.
Headlights and passing traffic. Car lights sweep across walls and ground and change large parts of the picture in a second.
Shadows and clouds. A cloud crossing the sun darkens the whole scene, and a moving shadow looks exactly like a moving object to a pixel counter.
Lighting and day and night switches. Floodlights switching on, or the camera flipping between colour and infrared mode at dusk and dawn, change almost every pixel at once.
Animals. Cats, dogs, birds and monkeys are real movement, and to basic motion detection a stray dog looks the same as a person.
The real cost is not the alarms themselves. It is what they teach the operator. When most alerts turn out to be leaves, the genuine one gets the same tired glance. Alarm fatigue is how a well equipped control room still misses an event.
Motion detection says that something in the picture changed. Motion tracking follows a moving object over time, frame after frame, and knows it is the same object throughout. AI detection adds the third piece: it names the object.
Motion detection gives you an event. Something changed in this zone at this time.
Motion tracking gives you a path. It is how a PTZ camera follows a subject, and how analytics such as line crossing know which direction someone went and loitering knows how long they have stayed.
AI detection gives you a class. Person, car, truck, bicycle, animal.
The three stack. The most useful alarms need all of them together: a person, crossing the fence line inwards, after hours.
Recording only when motion trips saves storage, often a great deal on quiet cameras such as a back corridor at night. It is the default on many consumer cameras.
It has two costs. First, motion only recording can miss the lead up to an event, and the moments before the trigger are often the most useful evidence. Second, if motion fails to trip, for example a slow figure in a dark corner at low sensitivity, nothing is recorded at all.
For cameras that matter, the stronger pattern is continuous recording, at a modest bitrate if storage is tight, with motion and AI events used to mark, search and keep footage longer. Keep motion only recording for cameras where a gap would not hurt.
Recording also brings duties. Footage of identifiable people is personal data under laws such as the UK GDPR and India's Digital Personal Data Protection Act, and the UK Information Commissioner's Office publishes detailed guidance on CCTV and video surveillance for organisations. A written retention period matters whichever way you record.
Most motion problems are fixed on site, not in software.
1. Start with placement. Keep trees, flags, roads and reflective surfaces out of frame wherever you can. It beats any setting.
2. Move the infrared away from the lens. On cameras that face heavy insect activity, an external infrared illuminator mounted a short distance from the camera keeps insects and webs out of the lens's glow. Clean housings on a schedule.
3. Draw zones. Watch the area that matters and ignore the rest: the pavement outside the fence, the neighbour's car park, the tree line.
4. Set the minimum size from a real walk test. Have someone walk the far edge of the zone at night, and set the minimum just below what they produce.
5. Set a maximum size where the system offers one, so a whole scene change such as lights switching on is not reported as motion.
6. Lower sensitivity until the noise stops, then repeat the walk test to make sure a person still trips it.
7. Use schedules. A loading bay that is busy all day needs motion alerts only after hours.
8. Add a cooldown, so one event produces one alert.
9. Review a week of alarms. Find the three biggest causes and fix them at the source.
There is no single best motion detection camera. For a business, the best choice is a good outdoor IP camera with a clean night picture and ONVIF support, paired with software that adds AI to its motion events.
For an intruder alarm indoors, a quality PIR or dual technology detector is still the best motion detector: cheap, reliable and frugal with power.
For an outdoor motion detection camera, judge the night picture first. A motion sensor camera with night vision is only as good as its night image, because noise in a dark picture becomes false motion. Our guide to night vision cameras covers infrared and low light choices in detail.
Then judge the system rather than the sensor. A consumer camera with a phone app suits a home. A business with dozens of cameras needs central zones, rules, AI and an audit trail, which an app does not provide.
Here is the honest position. For a staffed control room, raw motion is rarely the right alarm trigger. It is the right trigger for something cheaper: waking up the analysis.
Motion is the cheap trigger. It is lightweight, runs on every camera and says "look here".
AI is the decision. Intrusion detection checks whether the movement is a person, a vehicle or an animal, and whether it entered a protected zone, crossed a line in the wrong direction or lingered too long.
Rules are the response. A rule decides what happens next: an alert to the right team, an alarm in the operator console, a warning spoken through the camera speaker, a siren, or a message to another system. Rules also apply schedules, so the same person is welcome at two in the afternoon and an alarm at two in the morning.
Watch the camera itself too. A camera that has been covered, blinded or turned away produces no motion at all, so silence is not the same as safety. Tampering detection closes that gap.
Motion alone is still the right alarm in a few places: an empty, locked indoor space after hours, such as a server room or a vault corridor, where any movement is significant. It is also a sound recording trigger on a camera nobody watches live. Outdoors, and anywhere people work, put AI between motion and the operator.
Visylix is an automated AI video surveillance platform built in India. Motion detection is one of its 22 AI analytics, all built in house, and it is designed for exactly the role described above: a fast, cheap trigger that hands over to stronger analytics when pixels are not enough.
Motion on every camera. Lightweight motion analysis runs on every camera, with a sensitivity setting and a short cooldown so one event does not flood your alerts.
Zones that follow your day. Draw watch zones and ignore zones on each camera: ignore a road, a flag or a tree. Zones can run on a schedule, so a loading bay is watched only after hours.
Size limits in both directions. A minimum size ignores small changes such as insects, droplets and sensor noise. An optional maximum size means a whole scene change, such as lights switching on or the camera flipping to night mode, does not count as motion.
Each camera learns its normal. Visylix learns each camera's normal level of overall change, so a busy scene does not keep tripping a whole frame alarm.
Your cameras' own motion events as well. Visylix accepts motion events from the cameras themselves over ONVIF, normalised across brands, and drives them through the same rules.
AI takes over where pixels stop. Intrusion detection tells people, vehicles and animals apart. Line crossing knows direction. Loitering knows how long someone has stayed. Camera tampering detection catches a camera that is covered, blinded, blurred, moved or has lost its infrared.
Sudarshan acts. Every motion alert carries a snapshot. Through Sudarshan, the Visylix automation layer, motion and AI events can notify the right team, raise alarm incidents in the operator console, send signed webhooks with automatic retries, publish over MQTT, forward to your SIEM (security event monitoring) or PSIM, talk down an intruder through the camera speaker, fire siren and door strike relays, and start a recording clip of a length you set. A common pattern uses two rules: motion starts a clip, while the alarm waits for intrusion detection to confirm a person.
Your video stays with you. Visylix ships as a Docker image that runs on your own hardware: on premise, at the edge, hybrid or fully air gapped. Video never leaves the building.
Plans. Starter and Pro include unlimited cameras on the VMS licence, from ₹4,999 or $49 per month. AI analytics are a separate subscription billed per camera per month, so you add intrusion detection only on the cameras that need it. Talk to us for AI pricing.
Motion detection earns its place as a cheap, always on trigger on every camera; in empty, locked indoor spaces after hours, where any movement matters; for starting recording on cameras nobody watches live; for narrowing a search through hours of footage to the moments something moved; and as the first step in front of AI analytics.
It disappoints as a direct alarm outdoors, where trees, rain, insects, headlights and shadows outnumber intruders; on cameras with built in infrared and no cleaning routine; where sensitivity is turned up to make up for a poor night picture; as the only recording trigger on cameras that matter; and in any staffed control room where operators have learned to ignore it.
Motion detection in a security camera is software that compares each video frame with the frames before it and raises an event when enough of the picture changes. The event can start a recording, send an alert or trigger another system. It reports that something moved, but not what moved, so trees, rain and headlights can set it off as easily as a person.
A PIR sensor detects changes in heat as a warm body moves across its sensing zones. It has no picture, and typical models cover about 10 metres. Camera motion detection analyses the video itself, looking for changed pixels. PIR is less troubled by changes in light but sees no detail. Camera motion sees everything in the picture, which is why it also reacts to shadows, rain and headlights.
A plain motion sensor, such as the PIR detector in a burglar alarm or a sensor light, records nothing. It senses a change in heat and closes a switch. Some devices combine a motion sensor with a camera, and those can capture a photo or a video clip when triggered. If you are unsure, look up the model number, or ask whoever operates the device what it captures and keeps.
A motion sensor usually has a white or frosted plastic window, often faceted or dome shaped, with no lens visible behind it. A camera has a glass lens you can see into, and often a ring of small infrared lights that glow faint red in the dark. Combined units do exist, so looking up the model number is the reliable test.
Because pixels change for many reasons besides intruders. The usual causes are trees and flags in the wind, rain under infrared, insects and spider webs near the lens at night, headlights, moving shadows, and lights or day and night mode switching. Fix placement first, then draw ignore zones, set a minimum size from a real walk test, and lower sensitivity. Then let AI confirm what moved.
For a home or a single small shop, a camera maker's motion detection app can be enough. A business with many cameras, several brands or a control room needs more: central zones and settings, AI that tells people from animals, rules that send each alert to the right team, and an audit trail. That is the job of a video management system, not a phone app.
Rarely, on its own. Outdoors and in working areas, raw motion produces too many false alarms for operators to keep trusting it. Use motion to wake up AI analytics that confirm a person or a vehicle, and send the confirmed alarm. Motion alone is a sound alarm trigger in an empty, locked indoor space after hours, where any movement at all is significant.
Yes. Visylix runs lightweight motion analysis on every camera, with watch and ignore zones on a schedule, sensitivity, a cooldown, and minimum and maximum size limits. It also accepts motion events from the cameras themselves over ONVIF. When pixels are not enough, intrusion detection tells people, vehicles and animals apart, and Sudarshan rules alert, talk down, sound sirens or start recording. Everything runs on your own servers.