The parts of your system nobody explained.

Megapixels are a specification, not a result. What decides whether a recording is useful is pixels on the target — how much of the sensor lands on the thing you care about. The working number is pixels per foot (ppf): the camera’s horizontal pixel count divided by the width of the scene, in feet, at the distance you care about.

Design targets

The international video surveillance standard IEC 62676-4 sets operational requirement classes in pixels per metre — monitor, detect, observe, recognise and identify, in roughly a 1 : 2 : 5 : 10 : 20 progression from about 12.5 px/m to about 250 px/m. Converted to feet, those land near 4, 8, 19, 38 and 76 ppf, which is where the common North American shorthand of “about 40 to recognise someone you know, about 80 to identify a stranger” comes from. Treat them as design targets, not guarantees: lighting, lens quality, shutter speed, compression and how fast the subject is moving all move real performance.

The other three dials

• Frame rate. Most general surveillance runs comfortably in the low-to-mid teens of frames per second. Fast events — cash handling, gaming tables, vehicle lanes, high-speed lines — need more. Very low rates save storage and lose the motion that explains what happened.
• Bitrate. Set by resolution, frame rate, codec and how busy the scene is. H.265 and smart codecs cut bitrate hard on a static hallway and barely at all on a windy parking lot. There is no single correct number; measure it on your own scene if you can.
• Retention. Days of recording is bitrate multiplied by time — see the storage explainer below. Your retention target is often set by a regulator, a contract or an insurer rather than by preference.

At night, none of it matters as much as light. A low-light scene forces a longer exposure, which blurs anything moving; infrared gives you a monochrome image with hotspots near the lens. That is a lighting problem, not a camera problem.

Every electronic access control door is the same five-part chain, whatever the badge on the panel says.

Fail-safe versus fail-secure

These two words describe one thing only: what the locking hardware does when power is removed. Fail-safe hardware unlocks on power loss. Fail-secure hardware stays locked on power loss. Neither term says anything about people getting out, and mixing that up is how doors get dangerous.

Egress is a separate guarantee. From inside a space, a person must be able to leave — and on most doors that is mechanical: a lever, a panic bar, an exit device that retracts the latch no matter what the panel thinks. Fire-rated door assemblies have to latch positively, which generally rules out fail-safe electric strikes on them. Electromagnetic locks are inherently fail-safe and bring their own release requirements tied into the fire alarm. The governing documents are NFPA 101, the International Building Code, NFPA 80 and NFPA 72 — and section numbering moves between editions, so cite the edition your Authority Having Jurisdiction has adopted rather than a number you read online.

Two small inputs make the difference between a system that logs facts and one that logs noise: a door position switch, which tells you whether the door is actually closed, and request-to-exit, which tells the controller that an exit was expected so it does not record a forced door every time someone leaves.

An intrusion detection system has one job: notice that a protected space was entered, and tell somebody who will do something about it. Everything else is plumbing around those two halves — and the second half is the one people skip.

How it works

1. Points — individual sensors: door contacts, motion detectors, glass-break, vibration or shock sensors on walls and safes, beam sets across an aisle.
2. Zones and partitions — points grouped so a response means something. “Zone 7” tells nobody anything; “North dock roll-up” sends the officer to the right door.
3. Arming — the system is armed by keypad, credential, schedule or app. Perimeter-only arming lets staff work inside while the shell stays protected.
4. Entry and exit delay — a short window on designated doors so an authorised person can reach the keypad. Interior “follower” points inherit that delay; everything else trips instantly.
5. 24-hour points — panic buttons, tamper switches and supervisory points never disarm, whatever the schedule says.
6. Signal — the panel transmits over IP, cellular or both. Line supervision is what tells the monitoring centre that the path itself was cut.
7. Response — the monitoring centre verifies and dispatches per your written procedure and your call list.

False-alarm hygiene

Most jurisdictions have limited patience and a fee schedule to prove it, and many now require verification before dispatch. The causes are boringly consistent: a motion detector aimed at an HVAC diffuser or a hanging sign, door contacts with too much gap after the frame settles, an unbriefed cleaning crew, and user error — which is by a wide margin the leading cause. Control panels listed to the ANSI/SIA CP-01 false-alarm-reduction standard, enhanced call verification at the monitoring centre, clear zone names, real user training and a scheduled walk test do more for your dispatch record than any sensor upgrade.

Two more things that only matter when it counts: standby power sized so the system survives an outage, and a test record you can produce when an insurer or an auditor asks.

A modern security system is a network with cameras attached. Five constraints decide whether it behaves.

1. The PoE budget

Power over Ethernet comes in defined types. IEEE 802.3af (Type 1) supplies up to 15.4 W at the switch port, of which about 12.95 W reaches the device. 802.3at (Type 2, “PoE+”) supplies 30 W at the port and about 25.5 W at the device. 802.3bt adds Type 3 at 60 W and Type 4 at 90 W, delivering roughly 51 W and 71 W respectively. The trap is not the per-port figure but the aggregate budget: a 24-port switch rarely powers 24 devices at their maximum class. Pan-tilt-zoom cameras, heaters and infrared illuminators draw their peak at exactly the moment you need them — a cold, dark night.

2. The 100 metre channel

Balanced twisted-pair cabling under ANSI/TIA-568 is engineered as a 100 m channel: up to 90 m of permanent horizontal link plus up to 10 m of patch cords at both ends combined. Past that you move to fiber or a purpose-built extender, and you accept its limitations. This single number drives where the telecom rooms go — which is why IDF placement is a security decision, not just an IT one.

3. Segmentation

Put cameras and controllers on their own VLAN with no default route to the internet, and allow only the specific traffic the video management system needs. If your platform uses multicast, enable IGMP snooping so the stream does not flood every port. This is the cheapest security control on the whole project.

4. Switch and uplink sizing

Port count is the easy part. The number that fails is the uplink: every recorded stream on an access switch has to reach the recorder, and every live view pulled by an operator crosses it again. Add the camera bitrates, add viewing, and size the uplink with headroom rather than discovering it at 1 Gbps.

5. Power continuity

A camera on a UPS and a switch on utility power is a camera that is off. Back up the whole chain — switch, recorder, controller, and the network path to the monitoring centre. Runtime is a design decision driven by your generator strategy; batteries are consumables and get replaced on a schedule, not when they fail.

Storage sizing is arithmetic, and the arithmetic is easy. The hard part is the one input everybody guesses: bitrate.

Then adjust for reality

• Bitrate is scene-dependent. The same camera and settings produce a fraction of the data on a still corridor and a multiple of it on a windy car park with trees. Smart codecs widen that gap further. If the site exists, record a representative camera for a day and measure; if it does not, model a range rather than a point.
• Recording mode. Continuous recording is predictable. Motion or event recording is cheaper on paper and unpredictable in practice — a busy scene can trigger nearly continuously.
• Parity and formatting. RAID protection and file system overhead consume raw capacity. Size on usable capacity, and leave headroom — recorders behave badly at 100 % full.
• Extras that add up. Audio, a second lower-resolution stream for remote viewing, edge recording for failover and protected clips that are exempt from overwrite all take space.
• Retention may not be yours to choose. Regulation, contract or insurance often sets the number. Check before you design to 30 days out of habit.

One more: growth. Systems gain cameras, and a chassis that is exactly right on day one is short by year two. Size the enclosure and the drive bays for where the system is going.

Cable is the part of the system with the longest service life and the least attention. Done well it outlives three generations of camera; done badly it produces intermittent faults nobody can find.

The standards that govern it

• ANSI/TIA-568 — the cabling itself: categories, channel performance and the field test that proves it.
• ANSI/TIA-569 — pathways and spaces: conduit, cable tray, telecom rooms, sizing and bend radius.
• ANSI/TIA-606 — labelling and records, so somebody can find a circuit in five years.
• ANSI/TIA-607 — bonding and grounding of the telecommunications infrastructure.
• NFPA 70 (NEC), Articles 725 and 800 — the code side: cable listings, plenum and riser requirements, and the separation rules.

Conduit fill

The classic limits come from NEC Chapter 9, Table 1: a single conductor may occupy up to 53 % of the conduit’s internal area, two conductors up to 31 %, and three or more up to 40 %. Designers routinely stay below those ceilings for communications cable so a future pull is possible at all. Bends matter as much as fill — keep the total between pull points modest, respect the cable’s minimum bend radius, and pull with grease and patience rather than force. A cable stretched past its rated tension, or cinched hard under an over-tightened tie, fails its certification test and nobody can see why.

Separation from power

Communications and Class 2 circuits have to be kept away from power conductors, and the required separation depends on the code article, the voltage, and whether either run is in metallic raceway. Follow the edition your jurisdiction has adopted and the cable manufacturer’s guidance rather than a rule of thumb from a forum — and where a crossing is unavoidable, cross at right angles. Anything shared with a lighting circuit, a VFD or a large motor feeder is asking for noise you will chase for a year.

The unglamorous rest

Independent support — never lay cable on ceiling tile, sprinkler pipe or another trade’s hangers. Plenum-rated jacket anywhere in a return air path. Firestop every rated penetration, and photograph it before the wall closes. Label both ends, test every link, and hand over the test results as a deliverable.

Crime Prevention Through Environmental Design is the oldest and cheapest security technology there is: shaping a place so that legitimate use is easy and illegitimate use is awkward and visible. It is usually the highest-return line on a security budget, because it needs no power, no firmware and no licence renewal.

The four principles

1. Natural surveillance — lay out entries, windows, parking and landscaping so people can see and be seen. Trim the hedge that hides the back door.
2. Natural access control — use paths, fencing, planting, grade changes and lighting to guide people to the entrance you want them to use, before a card reader is involved.
3. Territorial reinforcement — make it obvious where public space ends and yours begins, with signage, surfacing, thresholds and maintained boundaries.
4. Maintenance and management — a broken fixture, a dead shrub and graffiti left up for a month all say the same thing: nobody is paying attention here.

Lighting for cameras, not for pavement

The instinct is to add brightness. The thing that actually helps is uniformity: the ratio between the brightest and darkest points in a scene. A pool of intense light next to deep shadow is worse than an even, moderate wash, because a camera exposes for one and loses the other, and a human eye arriving from the bright area cannot see into the dark one for several seconds.

Cameras also see people vertically, not from above. Illuminance measured on the ground says little about whether a face at the door will be legible; vertical illuminance at face height is the number that matters at entrances. Colour rendering matters too — a description of a jacket or a vehicle is only useful if the light lets colour survive. The Illuminating Engineering Society publishes recommended levels by application; they are design-specific, so a single “correct” figure does not exist and any consultant who offers one has not asked enough questions. Watch glare into camera lenses, light trespass onto neighbours, and local dark-sky ordinances.

Infrared illumination is a useful tool and a compromise: monochrome images, a bright hotspot near the lens, reflective surfaces that blow out, and a range that falls off faster than the datasheet suggests.

Three different jobs get bundled under one heading, and confusing them is how buildings end up with loud systems nobody can understand.

1. Alert — get attention. Tones, strobes, sirens. Fast, unambiguous, no information content.
2. Inform — say what is happening and what to do. Live or recorded voice, signage, desktop and mobile messages.
3. Communicate — two-way. Entry intercom, area-of-refuge stations, help points, officer radio.

Intelligibility is the requirement people skip

A voice system that cannot be understood is not a voice system. NFPA 72 addresses emergency communications and requires intelligible voice in the areas designated to receive it, evaluated with recognised speech-transmission measures rather than a sound level meter alone. Reverberant spaces — atria, gymnasia, warehouses, parking structures — are solved with more, smaller speakers at lower levels, not with fewer, louder ones. Turning up a system that echoes makes it less intelligible, not more. On Department of Defense projects, mass notification is governed by its own unified facilities criteria and is planned from the start rather than added later.

Intercom, and where it touches access control

Entry intercom is an access control device with a microphone: a door station calls an answering station, someone verifies who is there, and a release is granted. The egress rules from the access control explainer apply unchanged. Area-of-refuge communication is a life-safety requirement in many buildings, with its own supervision and answering obligations. Help points in parking areas are as much about visible reassurance as about calls placed.

Two design points that decide whether any of it works in an emergency: power and pathway survivability — the system has to keep running when the event is in progress — and who is authorised to press the button, written down and rehearsed before you need it.

Every camera, reader controller and recorder on your network is a small computer with a web server, an operating system and a login. Treated as an appliance, it becomes the softest target in the building.

The order to do it in

1. Change every default credential, uniquely per device, and store them somewhere a successor can find them. One shared “installer” password across a fleet is a single point of failure and an audit finding.
2. Separate accounts by purpose — the video platform’s service account is not a human administrator’s account. Remove the accounts your installer created for themselves at handover.
3. Run a firmware campaign, on a schedule, after checking vendor advisories. Firmware that is several years old is a published list of ways in.
4. Turn off what you do not use — cloud relay and peer-to-peer services, universal plug and play, unused discovery protocols, telnet and legacy remote shells, and the ports that go with them.
5. Never expose a device to the internet. No port forwarding to a recorder. Remote access goes through a VPN or the platform’s supported remote service.
6. Segment — a dedicated VLAN, deny outbound by default, and allow only what the system genuinely needs.
7. Synchronise time from a trusted source. A timestamp that drifts turns evidence into an argument.
8. Log and alert on configuration changes, failed logins and devices dropping offline. Keep the logs as long as you keep the video.
9. Wipe before disposal. A decommissioned recorder holds credentials, network topology and footage.

Know who actually made the device

Video surveillance hardware is widely relabelled, so the name on the bezel is not always the manufacturer of the board inside. For federal work this is not academic: Section 889 of the FY2019 National Defense Authorization Act restricts covered telecommunications and video surveillance equipment from named entities and their subsidiaries and affiliates, implemented in the FAR at 52.204-24, -25 and -26. Ask your supplier who builds the product and get the answer in writing before it ships — afterwards it is a rip-and-replace conversation.

A drawing set is not a picture of the system. It is a contract describing who buys what, who installs it and what “finished” means. Read it in this order and it stops being intimidating.

1. Cover sheet and sheet index — what exists, at what revision, and what the set is issued for. “Issued for permit” or “for review” is not “for construction.”
2. General notes and symbol legend — the legend defines every symbol on every later sheet. A symbol that is not in the legend is a question, not an assumption.
3. Site plan — the property, the approach, the exterior devices, the pathways coming into the building.
4. Floor plans by level — device locations, mounting heights, coverage, and the room names the schedules will refer to.
5. Enlarged plans — entries, lobbies, docks, telecom rooms and anything too dense to read at building scale.
6. Riser or one-line diagrams — how it connects: head end to IDF to device, power sources, and what rides on whose network.
7. Schedules — camera schedule, device schedule, door hardware schedule. The schedule, not the plan, says what the product is.
8. Details — the mounts, backboxes, penetrations and terminations, drawn because the words were not enough.
9. Specifications — CSI MasterFormat Division 27 for communications and Division 28 for electronic safety and security, with door hardware over in Division 08.

Four habits that prevent change orders

Find the order of precedence. The contract documents state which governs when the plan and the specification disagree — and eventually they will. Read the revision clouds. A delta triangle with a number ties to a revision note and a date; working from a superseded sheet is the most expensive mistake in the set. Never scale off a PDF. Dimensions and the schedule govern; a printed sheet may not be to scale at all. Hunt for “by others” and “NIC.” Conduit, backboxes, line voltage, firestop, patching and painting are frequently on someone else’s drawings, and the gap between two trades’ assumptions is where the money goes.

When something genuinely is not answered by the set, the answer is a written request for information, not a field decision. Field decisions are invisible until the closeout documents do not match the building.