All guides

Perimeter Intruder Detection Systems: Types and Costs

By Iu Ayala 13 min read Updated

A perimeter intruder detection system (PIDS), also called a perimeter intrusion detection system, raises an alarm when someone climbs, cuts or crosses a site boundary, before they reach a building. The main types are sensors on the fence, sensors buried in the ground, infrared or microwave beams, radar, and cameras with video analytics (software that spots people and vehicles in the picture). All of them give false alarms, so each needs a camera to show what set it off.

We build AI detection for existing CCTV, so this guide also says where cameras are the wrong tool.

The short version:

  • An ordinary yard or depot can usually use AI on its own cameras, placed to look along the fence. See which to choose, by type of site.
  • A long or remote boundary needs radar, thermal cameras or a fence sensor to detect, with a few cameras to verify. See how much fence one device watches.
  • Cameras and radar have public prices; fence and buried systems are quoted per site. See what it costs.

The types of perimeter intruder detection system

The UK's National Protective Security Authority (NPSA), the government body that tests these systems for critical sites, sorts them by where the sensor sits: on the barrier, in the ground, free-standing or wide-area. Cameras with analytics are not one of its PIDS categories, but they are common on commercial sites, so they are in the table.

Type and what it sensesWhat it needsWhat fools it
Fence-mounted: vibration from climbing or cutting the fenceA sound fence of a type the sensor was made forWind, rain, animals, debris, plants touching the fence
Electric fence: a live wire cut or earthedA host fence strong enough to carry it, out of the public's reachVegetation and anything that bridges the wires
Buried: footsteps and vehicles over a hidden cableA trench, and ground nobody walks on by dayRunning water, fast-draining soil, tree roots, animals
Beams: an infrared or microwave beam being brokenA straight, clear line of sight; 60 to 200 m a pair by modelFog and heavy rain (infrared), animals, swaying plants
Radar: anything moving across open groundOpen ground: walls and vehicles cast shadowsMetal fences and parked lorries (reflections), swaying trees
Camera analytics: a person or vehicle crossing a line drawn on ordinary CCTVLight or infrared, and the person within about 20 to 30 mDistance, fog, glare, insects on the lens
Thermal camera analytics: the heat of a person or vehicleA clear view; no light at allDense fog and heavy snow

Two things in that table matter more than the rest. Fence and beam sensors tell you that something happened and roughly where. Only a camera tells you what. And nothing in the table is free of false alarms, which is the subject of the next section.

Fence-mounted sensors

A sensing cable is tied along the fence, or small vibration sensors are fixed to it at intervals, and an analyser listens for the pattern of someone climbing or cutting. The cable can be fibre optic, which NPSA says is common on sites with long perimeters. NPSA calls these the most common barrier-mounted type. They only work as well as the fence: NPSA's guidance says a fence that is too rigid will not pass the vibration on, and one that is too loose will alarm in wind and rain. A rattling chain-link fence with ivy on it is a poor host. Welded mesh in good repair is a good one.

Buried sensors

Sensors under the surface pick up people and vehicles crossing above. NPSA lists three kinds: vibration cables that feel footsteps, fluid-filled tubes that feel pressure, and "leaky feeder" cables that set up a radio field and notice a body moving through it. Their advantage is that nothing shows. NPSA calls them "invisible to intruders" and hard to defeat without inside knowledge, which matters on a site worried about being scouted. They also follow dips and slopes that beams cannot.

The drawbacks are the groundworks and the fact that a buried sensor cannot be moved. NPSA advises putting them in a sterile zone, a strip between two fences that nobody should be in, because staff, deliveries and animals crossing the line all look like intruders. Few working yards have a strip like that, so buried systems suit sites that do: utilities, solar farms and large estates.

Beams

A transmitter on one post sends infrared beams to a receiver on another, and breaking them raises the alarm. Outdoor sets are sold by range: Optex's SL-350QDM, a common four-beam model, covers 100 m. Microwave links do the same job with radio waves and, NPSA notes, cope better with bad weather and uneven ground than infrared; laser-based detectors draw the same kind of invisible barrier. Both need a straight run with nothing growing into it, microwave links have dead spots near each post, and neither can tell a person from a deer. Outdoor PIR detectors, which sense moving body heat, are the cheaper free-standing option for short distances, though NPSA says that outdoors they "generally suffer from poor detection and high false alarm rates"; our section on PIR sensors covers what fools them.

Radar

A security radar watches an area, not a line. The Axis D2110-VE, a widely stocked model, detects a person from 3 to 60 m across a 180° sweep according to its manual. It needs no light, and Axis describes it as built for "various weather and all light conditions". It reports where the person is and can steer a moving camera onto them. The same manual lists its weak points: any wall, fence, tree or large bush leaves a "radar shadow" behind it, and metal fences cause reflections that show up as ghost tracks. Radar suits open yards and fields. It is a poor fit for a yard stacked with containers.

Cameras with video analytics

The camera, the recorder or a separate computer runs AI that recognises people and vehicles, and you draw a line or a zone on the picture. Crossing it raises the alarm, with the video clip attached. This is level two of the four levels of AI in security cameras. Our guide to CCTV analytics covers the rule types and how far to trust each. The limit is distance: a person far from the camera is a few dozen pixels tall, and the AI stops being sure what it is looking at.

Thermal cameras see heat instead of light, so they need no lighting and reach further. Axis's white paper on thermal cameras says that in moderate fog or smoke they "detect objects that would be completely invisible to a visual camera". Axis's datasheet for its Q1961-TE works out that it can recognise a person at 50 m with a 7mm lens and 98 m with a 13mm lens. Those are figures on paper, and Axis says to check them on site. Thermal cameras cost several times as much as an ordinary camera and show a glowing shape, not a face, so you still want an ordinary camera for evidence.

Every type gives false alarms, so every type needs a camera

NPSA is blunt about this. Its guidance says "no PIDS would achieve zero false alarms", that an acceptable rate is 5 false alarms a day per kilometre of perimeter in good weather, and that the rate rises in bad weather. That yardstick is for systems good enough to protect power stations. On a 200 m boundary that is one false alarm a day, or 7 in a calm week.

So NPSA's rule is that every alarm is checked on CCTV before anyone responds, and that the perimeter is split into zones with CCTV and lighting planned for each. In practice a perimeter system is always two things: something that detects, and a camera that verifies. So the choice is what does the detecting, because you need the cameras either way.

Ask how the two are joined. An alarm from a fence zone or a beam should bring up the camera for that stretch on the screen by itself, with the clip from the seconds before, so nobody is hunting through camera views at 2am. A recorder, or the video management software that runs a larger camera system, does this when the sensor is wired or networked into it. Radar goes a step further: it knows where the person is, so it can steer a camera that turns and zooms to follow them.

For camera analytics, false alarms come from the things our guide to motion detection false alarms covers: insects on the lens at night, headlights, rain. Person detection removes most of them. It does not remove a fox at 25 m that the AI mistakes for a crouching person.

How much fence one device watches

This is where the types differ most, and where quotes go wrong. A camera used for detection looks along the fence, not at it, so that an intruder has to cross its view. It cannot see the ground right under itself, and beyond a certain distance the AI can no longer tell a person from anything else. Hikvision's guidance for its AcuSense cameras puts the blind zone at 3.7 m for a wide 2.8mm lens mounted 3.3 m up, and 5.6 m for a narrower 6mm lens, which sees further but covers less width. The same guidance recommends no more than 10 m for the wide lens and 22 m for the 6mm on older models. Hikvision's 2020 generation raised the distance at which it tells a person from anything else to 20, 25 or 30 m depending on the model. We use 20 m and 30 m here, the best case; the AcuSense guide has the detail.

Each camera's blind zone has to be watched by the camera behind it. So a wide-lens camera adds 16.3 m of watched fence and a 6mm camera adds 24.4 m. A radar set 15 m back from the fence watches about 116 m of it. Drag the slider to the length of your own longest fence line.

How many devices it takes to watch a straight fence line. Each AI camera, 2.8mm lens adds 16.3 metres of watched fence, each AI camera, 6mm lens adds 24.4 metres and each radar about 116 metres. A 200 metre fence takes 13 wide-lens cameras, 8 cameras with 6mm lenses or 2 radars.

AI camera, 2.8mm lens sees a person from 3.7 m to 20 m

8 cameras

AI camera, 6mm lens sees a person from 5.6 m to 30 m

5 cameras

Radar sees a person up to 60 m away, set 15 m back

2 radars

20 m400 m
  • Fence watched; each shade is one device
  • Not watched: under the first camera
The dashed line is the fence and each dot is a device. Lengths along the fence are to scale; the depth of each shape is not. Blind zones are from Hikvision's AcuSense guidance (3.3 m high, tilted 10°); the 20 m and 30 m ranges are its 2020-generation figures, matched to lens by us. Radar figures are Axis's for the D2110-VE. A straight, clear fence line is assumed: every corner and obstruction adds a device.

A 200 m fence line takes 13 wide-lens cameras, 8 with 6mm lenses, or 2 radars plus a camera or two to verify. With ordinary AI cameras, plan on one every 16 to 24 m at best.

A fence-mounted sensor works differently: the cable runs the whole length, and one analyser listens to a long stretch of it. Senstar's FlexZone, for example, covers up to 600 m per processor and places an intrusion to within 3 m. The installer then splits that length into zones, each with a camera to verify.

That sounds like an argument against cameras. For a long, empty boundary it is. Past a few hundred metres, the other detectors need far fewer devices: a radar watches about 116 m, a thermal camera can recognise a person at 50 to 98 m on Axis's figures, and one fence-sensor processor covers up to 600 m. A few ordinary cameras then verify. But most yards do not need the whole fence watched to the same standard. Intruders come through the gate, over the stretch that backs onto a lane, or at the corner nobody can see from the road. Watching those properly, with cameras you mostly already own, beats a thin layer of detection spread along every metre.

Can the CCTV you already have be the perimeter system?

Often, yes, for a site with up to a few hundred metres of boundary. It works when:

  • The cameras look along the boundary. A camera on the building looking out at the fence square-on sees an intruder as a small figure for a second or two. One looking along the fence sees them cross its whole view.
  • The far end is within range. Use the figure above. Past the range, add a camera or change the lens; turning the sensitivity up only adds false alarms.
  • There is light. A camera's own infrared lamps fade with distance, so a dark far corner needs a lamp or a thermal camera.
  • The AI is person detection, not plain motion. Many Hikvision and Dahua cameras sold in the last few years have it; older ones can get it from a new recorder or a separate AI box. Our AI CCTV guide walks through the options.
  • Somebody receives the alarm. A phone alert to an owner who is asleep is not a response. Decide who looks, and what they do, before buying anything.

The alarm rule matters as much as the hardware. Draw the line a metre or two inside the fence, not on it, so that people walking past outside do not trigger it. Arm it on a schedule. Where the built-in rules are not enough, custom detection can go further. For a UK CCTV security company we built real-time trespass and violence detection that runs on a computer at the customer's site, 15 to 20 cameras per graphics card, with the footage never leaving the building.

Cameras are the wrong detector when the boundary is long and remote, when there is no power or network along it, when fog is common, or when you need to know about someone touching the fence from the outside before they are over it. That last one is what fence-mounted sensors are for.

What installing and maintaining each type involves

  • Fence-mounted: fitted to the inside of the fence, where an intruder cannot reach it, and split into zones so an alarm names a stretch. NPSA's warning is to match the sensor to the fence type. Upkeep is keeping plants off the fence and retuning as it loosens with age.
  • Buried: a trench along the whole line, in the kind of ground the maker specifies; NPSA says the wrong ground raises false alarms and lowers detection. It is the most disruptive to install and the hardest to change later.
  • Beams: posts at each end of every straight run, aligned to each other, on flat ground. Microwave links are overlapped to cover the dead spot at each post. Upkeep is cutting back anything that grows into the line and realigning posts that move.
  • Radar: one device on a pole or wall with a network cable. Axis recommends 3.5 to 4 m high for its D2110-VE. Setting up means drawing zones and excluding metal fences and trees that cause false tracks.
  • Cameras: a network cable to each, which also powers it. Hikvision's guidance for person detection is 3 to 5 m high, tilted down about 10°. Upkeep is cleaning lenses and cobwebs, and redrawing zones when the yard layout changes.

Whatever detects, NPSA recommends a backup power supply, because a perimeter system that goes down in a power cut is easy to defeat. Adding a camera later is a small job. Extending a fence sensor or burying more cable means setting up and testing again every zone it touches.

What a perimeter intruder detection system costs

Only some of this market publishes prices. Cameras and radar are sold through retailers; fence-mounted and buried systems are quoted per site after a survey, and we could not find a UK supplier that publishes a price per metre. Prices below are UK retail including VAT in November 2026, with dollar figures at £1 = $1.34. Poles, cabling, power and labour are extra, and on a long fence they can cost more than the devices.

DetectorPriceFor a 200 m fence line
AI camera, 4MP (Hikvision AcuSense turret)£100 to £135 ($135 to $180) each8 to 13 cameras: £800 to £1,755
Security radar (Axis D2110-VE)From about £1,555 ($2,085) each2 radars: about £3,110, plus cameras to verify
Infrared beamsBy quote; one pair per straight run2 pairs of 100 m beams, if the line is straight
Fence-mounted or buried sensorsBy quote, after a site surveyZones set by the installer, each with a camera

When a supplier quotes a price per metre, ask what it leaves out. It is usually the sensor only: the analysers, the power to reach them, the cameras and lighting for each zone and the visits to set it up and tune it are separate lines. Ask too what it costs each year.

Our own work is custom AI on cameras a site already has: a £4,000 to £6,000 feasibility study on your footage, credited to a £15,000 to £20,000 build. It is not the cheap route for a standard yard, and we would not sell it as one. The built-in person detection on a £100 camera handles a line-crossing rule well. Custom work earns its cost when the built-in rules keep getting it wrong on your site, when one system has to cover many sites, or when you need to detect something the camera makers do not offer.

Which to choose, by type of site

SiteDetect withWhy
Yard, depot or builders' merchant, under about 300 m of fenceAI cameras along the fence lines and on the gateThe cameras are mostly there already, and every alarm comes with video
Construction siteTemporary camera towers with person detection, monitoredThe boundary moves, and there is nothing fixed to mount sensors on
Open storage, car compound or fieldRadar, with a moving camera it can steerOne device watches a wide open area in the dark and in most weather
Solar farm or long rural boundaryFence-mounted sensors or thermal camerasToo long for ordinary cameras, no lighting, nobody nearby
Utility, data centre or other critical siteAn NPSA-evaluated PIDS, designed by a specialistInsurers and regulators may require it; this guide is not enough

Our advice for the first row, which is most businesses: fix the camera positions before buying a sensor. Move or add cameras so they look along the fence, switch on person detection with a line inside the boundary, and run it for a month while counting the alarms. If it is quiet and catches your own test walks, you have a perimeter system. If one stretch keeps failing, you now know exactly which stretch needs a radar or a fence sensor.

Monitoring and police response

Detection is worth little if nobody acts on it. The usual arrangement in the UK is a monitoring centre whose operators see each alarm clip, speak through a loudspeaker on site and call a keyholder or the police. The loudspeaker is there so that the intruder leaves before anyone has to attend.

For the police to attend, the system needs a unique reference number (URN) from the local force, and it must be installed and maintained to BS 8418, the British Standard for detector-activated CCTV (temporary camera towers have their own code, IA 2401), by a company certified by an accredited body such as NSI or SSAIB. The National Police Chiefs' Council requirements for England, Wales and Northern Ireland withdraw police response after three false calls in a rolling 12 months. A person seen on site with no crime in progress is treated as trespass, a civil matter, and counts as a false call. This is why a noisy perimeter system is worse than none: it spends the three calls on people who turn out to have done nothing.

Cameras and radar that look beyond your fence also record people who are not on your land. That is allowed with a good reason, but it brings duties under UK GDPR, which our section on filming the street explains. US rules vary by state.

How to test a perimeter system before you rely on it

NPSA's evaluation scheme typically tests a system for eight to nine months, with attack trials for the detection rate and false alarm trials for the false alarm rate. You can run a small version yourself:

  • Walk, run and crawl across the line at the near end, the middle and the far end of each camera's or sensor's stretch. Crawling is the hardest of the three for camera AI.
  • Do it after dark, and once in rain. A system that passes at 3pm proves little.
  • Wear dark clothes and try the corners. Corners and gateposts are where zones meet and gaps hide.
  • Count false alarms for two weeks before connecting anything to a monitoring centre. Our rule of thumb: more than one or two a night and whoever receives them starts to ignore them.
  • Time it. From crossing the line to a person seeing the clip should be seconds. The weapon detection we built for Angel Protection alerts in about a second across more than 1,000 cameras; a perimeter alarm that arrives a minute later arrives after the intruder has gone inside.

If you would like the cameras you have tested this way, that is what our feasibility study on your own footage does, and the Angel Protection case study shows the same approach at scale.

Questions

What is a perimeter intrusion detection system (PIDS)?

It is a set of sensors on or near a site boundary that raises an alarm when someone climbs, cuts or crosses it. It is often shortened to PIDS, and "perimeter intrusion detection system" means the same thing. It detects; the fence delays; cameras verify.

What is the difference between a PIDS and CCTV?

A PIDS senses a physical event, such as vibration on the fence or a broken beam, and reports a location. CCTV records pictures. CCTV with AI analytics can also detect, within about 20 to 30 m of each camera. High-security sites use both; many commercial yards use AI cameras alone.

What is the difference between a perimeter alarm and an intruder alarm?

An intruder alarm protects the building: door contacts and motion sensors inside. A perimeter system works outside, at the boundary, so the alarm comes while the intruder is still in the open.

How far can a perimeter camera detect a person?

Hikvision rates its 2020 generation of AI cameras to tell a person from anything else at 20 to 30 m, depending on the model; older ones manage 10 to 22 m. Thermal cameras and radar reach further: the Axis radar above detects a person at 60 m.

Do perimeter detection systems work in fog and rain?

Radar and microwave are the least affected. Fence sensors work but alarm more in wind and heavy rain. Infrared beams and ordinary cameras lose range in fog. Thermal cameras cope with moderate fog but not dense fog.

Can a perimeter system tell an animal from a person?

Fence sensors and beams cannot. Radar can ignore small objects such as cats and rabbits, and AI cameras classify people and vehicles and ignore most animals. Deer and large dogs still cause some alarms on every type.

Do I need planning permission for a perimeter intruder detection system?

Not normally for the sensors. The fence is another matter: in England a new fence over 2 m high, or over 1 m next to a road, needs planning permission according to the Planning Portal. Camera poles and listed buildings can need it too, so check with the council.

Is a wireless perimeter system any good?

Battery beams and wireless sensors suit temporary sites and places with no cabling. They are easier to install, but the batteries need changing. For a permanent site, wired power and network are more dependable and let you add cameras, which you need anyway.