Commercial security camera system design for usable footage

Most footage from a commercial security camera system is reviewed once, under pressure, and found to be useless. The camera saw the incident, but the face is eight pixels wide, the plate is a smear, and the angle that mattered was never covered. That is decided at design, not at playback: coverage was guessed from a floor plan and resolution at distance was never calculated.

Pixels per metre decides what the footage is worth

Whether footage identifies a person or merely proves someone was there comes down to pixels per metre: how many horizontal pixels the camera lays across one metre of scene, at the distance the subject stands. The trade works to four bands, labelled detection, observation, recognition and identification.

Pixels per metre falls off with distance, because a fixed lens spreads the same pixel count across a wider scene. A 4K camera covering a scene 20 m wide lays 3840 pixels across those 20 m, roughly 190 px/m; at twice the distance the same lens covers about 40 m and gives roughly 95 px/m. So the question is never how many cameras, but what each one has to answer, from where, at what distance.

Lens and mounting height beat megapixel count

Sensor resolution sets a ceiling; the lens decides how it is spent. Halving the field of view doubles pixels per metre on the target, usually costs less than doubling sensor resolution, and avoids the low-light penalty of a denser sensor.

Mounting height works the same way. A camera at 3.5 m in a ceiling corner gives an excellent view of the tops of heads; at 2.4 m, on the approach side and facing the direction of travel, it gives faces.

Camera placement for entrances, parkades, loading and amenity areas

Placement is a set of specific jobs, not blanket coverage. Each area has one question it needs answered.

• Entrances and lobbies. One identification camera on the approach side, low enough to see faces and aimed along the direction of travel, plus an overview camera for context. Shooting outward through lobby glass renders faces as silhouettes, and repositioning so faces are lit does more than wide dynamic range.
• Parkades. Plate capture needs its own camera on a tight lens aimed along the vehicle path, exposed for headlights and retroreflective plates; an overview camera will not read plates at any resolution. Aisle cameras cover movement, and slab heights, columns and mechanical services constrain where they can go.
• Loading bays and service doors. Goods leave here and unauthorized entry is easiest here, so it earns identification-grade coverage both ways: one camera inward at the door, one on the apron. Housings take weather, vibration and wash-down.
• Amenity rooms and parcel storage. A parcel room justifies one identification-grade camera on the door rather than three wide cameras over the shelving; the requirement is knowing who came in.

Where cameras should not point

Coverage design includes what stays out of frame: washrooms and change rooms, a suite interior visible through an open door, and workspaces where staff would sit under continuous observation. Recording in a residential or workplace setting in British Columbia carries privacy obligations around purpose, notice, access and retention, so the plan should be reviewed with the strata council, the employer, or legal counsel before it is built.

Low light, infrared and what defeats a camera at night

Most unusable footage is night footage. Lens aperture and sensor size do more after dark than resolution does, and a lower-resolution camera with a fast lens often out-performs a higher-resolution one in the same position.

Built-in infrared has a working range, and past it the scene falls to black. Closer in the opposite happens: IR reflects off walls, columns, rain and falling snow, and the returned light washes out the sensor, so a dome in a tight parkade corner can blind itself. IR is also monochrome, so if clothing or vehicle colour matters at night the answer is site lighting and a low-light colour camera, not a stronger IR array.

Motion is the other trap. Freezing a walking person needs a short shutter, which needs light, so clean night images are often bought by slowing the shutter and blurring exactly the motion you care about. Fixing the lighting is regularly the cheapest option available.

PoE, the 100 m limit and where the switches land

Cameras take power and data over one cable, and the limit on that cable dictates where the communications rooms go. A balanced twisted-pair horizontal channel is limited to 100 m including the patch cords at both ends, leaving roughly 90 m for the permanent link.

Distance is measured as the cable runs, not as the drawing scales. A parkade run that looks like 80 m on plan can pass 100 m once tray routing, vertical drops, service loops and the route around a shear wall are counted; the answer then is a switch closer to the cameras, fed with fiber.

Two things then decide the switch. PoE budget is a per-switch number, not per-port, so an eight-port switch may not simultaneously power eight exterior cameras drawing for heaters, IR and motors. And since camera power comes from the switch, recording continues only while that switch is up, which puts switch and recorder on UPS. Camera drops belong in the same structured cabling design as data, access control and Wi-Fi.

Bandwidth on a shared network

Camera traffic is continuous rather than bursty, which is what separates it from office traffic. Keep the recorder on the same switch or stack as its cameras so the heavy path never crosses the building uplink, and put cameras on their own VLAN. Remote viewing runs the other way, over the site's upload bandwidth — usually the smaller number — so apps should pull substreams.

NVR or VMS, and sizing retention for a commercial security camera system

An NVR is an appliance with fixed capacity running its manufacturer's cameras; a VMS is software on a server that scales across buildings and integrates with other systems.

For a single building with a settled camera count an NVR is usually right, and a VMS is not worth paying for. A VMS earns its licence cost across a growing portfolio, a mixed camera fleet, or integration with access control.

Sizing retention

Retention is one calculation: cameras × bitrate per camera × days. Change any of the three and storage moves proportionally, so the honest conversation is about which one you reduce.

Bitrate is set by resolution, frame rate, codec and scene activity. H.265 compresses better than H.264 on the same image, and a still corridor consumes a fraction of what a busy street does. As arithmetic, a stream held at 1 Mbps continuously is about 10.8 GB per day, and the total scales from there.

Two things reliably break a storage estimate:

• Nominal bitrate is not average bitrate. It rises with motion, weather and image noise, so a snowy night with IR on can push cameras to their ceiling and overwrite the days you meant to keep. A per-camera maximum bitrate turns an unpredictable estimate into a bounded one, at some cost in quality during busy scenes.
• Drives must be rated for continuous write. Surveillance is a sustained-write workload and desktop drives fail early under it.

Remote viewing and access control integration, done securely

Do not forward ports to a recorder. It puts a device with an administrative web interface and a long history of firmware vulnerabilities on the public internet, where automated scanning finds it. Three paths cover almost every requirement: the manufacturer's outbound relay service, where no inbound port opens but you should confirm where video is relayed and who holds the account; a VPN to the site, which keeps the recorder private but needs a managed firewall and per-user provisioning; or a hosted VMS connecting outbound only.

The basics apply on every path: individual accounts rather than a shared admin login, credentials removed when staff leave, default passwords changed at commissioning, and firmware kept current.

Making an entry event pull the right camera

Integrating surveillance with access control means a credential presented at a door produces the video of that door at that second, rather than a search through hours of recording. Three things settle it at design: both systems on one time source, since clock drift between controller and recorder makes correlation worthless; a documented map of which camera covers which door; and a supported integration path, whether a shared vendor platform, a VMS module, or an event bridge.

Done properly, a card at a loading door bookmarks the recording and opens the camera that saw it, so whether the person who entered is the person on the credential is answered in seconds.

What to do next

Start with a list of questions, not a camera count. For each point in the building, write down what the camera has to prove, what the subject is, how far away it stands, and what the light is doing there at 3 a.m. That list converts directly into lens selection, mounting height and a defensible pixels-per-metre target.

Then check the two items that quietly reshape budgets: cable distances against the 100 m limit, which decides where switches go, and retention against camera count and bitrate, which decides the recorder. Confirm the privacy review before drawings are finalized. If you want coverage, cabling and retention scoped in one pass against the actual building, Orbit Automation offers a site assessment across Metro Vancouver and the Fraser Valley.