Three metrics, not one#
A perimeter intrusion detection system exists to give a response force enough warning to act. Three metrics describe whether it does:
- Probability of detection for a defined intruder and approach — always conditional on the threat, never a single brochure number.
- Nuisance alarm rate — alarms from wind, animals and weather, per zone per unit time.
- Probability of interruption — whether detection happens early enough, given delays and response time.
Most failed perimeters scored well on the first metric in an acceptance test and badly on the other two in operation.
Sensor families#
| Sensor | Detects | Strengths | Typical failure mode |
|---|---|---|---|
| Fence vibration / taut-wire | Climbing, cutting | Precise location on the fence line | Wind and rain on the fence |
| Buried fibre / pressure | Footsteps, vehicles | Covert, follows terrain | Ground saturation, frost, traffic |
| Ground surveillance radar | Moving people, vehicles | Wide area, all weather, 360° | Vegetation movement, birds, multipath |
| Thermal imaging | Heat contrast | Night detection and classification | Thermal crossover at dawn/dusk, heavy rain |
| Video analytics | Motion, classes | Identification, evidence | Lighting, shadows, insects on lens |
| Elevated sentries | Anything in view | Raise every sensor above clutter; movable | Need perches; weather for transit |
Independence matters more than count. Two sensors driven by the same nuisance source — wind shaking a fence and the shrubs in front of a radar — add less than their specifications suggest.
The Pd / NAR trade-off, quantified#
Fusing two sensors with OR logic raises detection and adds nuisance rates; AND logic does the opposite:
In our operator-load model a 40-zone perimeter with OR logic produced 76 alerts an hour on a windy night against an operator capacity of about 20; AND logic cut that to 0.4 but missed one intruder in seven. Automated visual verification of OR alerts kept detection at 96.5% with 2.28 alerts an hour.
Detect early or don't bother#
The EASI-style model in probability of interruption shows why fence-line detection is often too late: once an intruder is over the fence, the remaining path delay at a typical unstaffed site is a few minutes, while response times are often ten or more. Moving detection out to the approach added about 150 s of warning and lifted at a 10-minute response from 0.37 to 0.62 in our example. Extra sensors inside the fence changed it by less than 0.01.
Placement#
Coverage is a maximum-coverage problem with a submodular objective, so greedy placement is near-optimal in practice and provably within of optimal. In our viewshed study on 30 synthetic sites, raising a sensor from 2 m to 9 m doubled visible ground (13.9% → 28.3%), and two greedily placed masts matched six random ones. Survey, then place.
Standards and regulation#
Utilities in North America design substation security under NERC CIP-014, which requires transmission owners to identify critical transmission stations and substations, assess threats and vulnerabilities and implement physical security plans; a revised CIP-014-4 was approved by FERC in September 2026. Camera resolution requirements are commonly specified with IEC 62676-4 DORI densities (our guide). For drone threats to perimeters, see the counter-drone briefing and radar vs. RF detection.
Worked example: a 1.2 km substation perimeter#
Consider a transmission substation with a 1.2 km fence divided into 24 zones of 50 m, an unstaffed site with a mean police or security response of 10 minutes, and an operations center that already monitors 30 other sites.
Alarm budget. The center has two operators on nights. If this site may use at most 10% of their capacity, its budget is about alerts per hour on the worst nights. Fence and radar fused with OR logic at our windy-night rates would produce alerts per hour — over ten times the budget. With automated visual verification the same perimeter produces about . Budget met.
Timeliness. With a 10-minute response and roughly 11.5 minutes of path delay after the fence, fence-line detection gives a probability of interruption around 0.37 in our model. Detecting on the 300 m approach raises it to about 0.62; hardening the control-house door adds more. The design therefore places a radar mast to cover approaches and a perched sentry for verification and deterrence, rather than adding fence sensors.
Placement. A viewshed run on the site's survey data picks the mast and perch positions greedily; the expected outcome, from our synthetic study, is that two well-placed elevated sensors match the coverage of several more placed by convenience.
Resolution. The sentry's camera must reach recognition density (125 px/m) at the far edge of its zone; with a 4K sensor and a 30° field of view that holds to about (DORI).
The resulting system costs less than adding sensors everywhere and performs better on the two metrics that decide outcomes — operator load and timeliness.
A design sequence that works#
- Define the threat and the asset, then walk the likely paths.
- Measure response time honestly, including night and weekend.
- Set an alarm budget per operator-hour and design fusion to meet it on bad-weather nights.
- Place detection to maximize , not fence coverage.
- Survey line of sight; place sensors greedily; check resolution against DORI.
- Baseline for 30–90 days before tuning.
Frequently asked questions#
What is a perimeter intrusion detection system?
A set of sensors, fusion logic and assessment tools that detect people or vehicles crossing a defined boundary early enough for a response.
What is a good probability of detection?
It depends on the threat definition. Specify it per intruder type and approach, and evaluate it together with nuisance alarm rate and probability of interruption.
How do you reduce perimeter false alarms?
Fuse sensors with independent failure modes, add automated visual verification before alerts reach operators, and tune during a baseline period that includes bad weather.
References#
- M. L. Garcia, Design and Evaluation of Physical Protection Systems, 2nd ed., Butterworth-Heinemann / Elsevier, 2007.
- NERC Reliability Standard CIP-014 (Physical Security) — transmission stations and substations; revised CIP-014-4 approved by FERC in September 2026. nerc.com
- IEC 62676-4:2014, Video surveillance systems for use in security applications — Part 4: Application guidelines. DORI pixel densities as summarised in Axis Communications, “Pixel density based on IEC 62676-4:2014.” whitepapers.axis.com