The question we are asked most often is "how many cameras do we need for a 5,000 square foot office?" It is an entirely reasonable question, and it has no useful answer.
Camera count does not follow from floor area. It follows from what you need to be able to prove, and where.
Start with the three levels of detail
Surveillance design distinguishes between three fundamentally different outcomes. Each requires a different pixel density on the target, and therefore a different combination of resolution, lens and mounting distance.
Detection — "something is happening there"
You can tell a person is present and moving. You cannot tell who. Suitable for perimeters, car parks and large open areas where you want awareness of activity. Roughly 25 pixels per metre of scene width.
Recognition — "that is someone I know"
You can recognise a person you already know — an employee, a regular visitor. Not enough to identify a stranger to police. Suitable for corridors, general work areas and reception approaches. Roughly 125 pixels per metre.
Identification — "that is evidentially this person"
Facial detail sufficient to identify a stranger with confidence. This is what you need at every point where evidence actually matters. Roughly 250 pixels per metre or more.
The single most common CCTV failure is installing twenty detection-grade cameras and then discovering, after an incident, that not one of them produced identification-grade footage of the door the person came through.
Where identification-grade coverage belongs
Every choke point — the places everyone must pass through:
- Main entrance and every secondary entrance, at face height, angled to capture faces on approach rather than the tops of heads
- Reception and visitor desk
- Cash handling points and tills
- Server room, stock room and any restricted area door
- The gate or vehicle entry point, with a separate camera for number plates
Everywhere else, detection or recognition grade is usually sufficient and considerably cheaper.
Camera placement mistakes we see constantly
Mounting too high
A camera at four metres looking down at 45 degrees captures the top of a head and a pair of shoulders. Entrance cameras intended for identification should be mounted at roughly 2.4–2.8 m and angled close to horizontal along the direction of approach.
Pointing at a window or open door
Backlighting turns every subject into a silhouette. Either reposition the camera or specify one with genuine wide dynamic range — and verify the WDR figure rather than trusting the marketing claim.
One wide camera doing several jobs
A single wide-angle camera covering an entire floor gives you a scene, not evidence. Two well-placed cameras with appropriate lenses almost always beat one wide camera trying to cover everything.
Ignoring night conditions
Walk the site after dark before finalising positions. IR illuminators have a real, limited range, and IR reflecting off a nearby wall or a glass pane washes out the entire frame.
Sizing storage honestly
Storage requirement is driven by camera count, resolution, frame rate, codec and whether you record continuously or on motion.
As an indicative figure: a 4 MP camera recording continuously at 15 fps with H.265 compression consumes roughly 15–25 GB per day, depending on scene complexity. A busy scene with constant movement sits at the top of that range; a static corridor sits near the bottom.
| Cameras | Retention | Indicative raw storage |
|---|---|---|
| 8 × 4 MP | 30 days | ≈ 4–6 TB |
| 16 × 4 MP | 30 days | ≈ 7–12 TB |
| 32 × 4 MP | 30 days | ≈ 14–24 TB |
| 32 × 4 MP | 90 days | ≈ 43–72 TB |
Three things people forget:
- RAID overhead. RAID 5 or 6 costs you the capacity of one or two drives. Usable capacity is not raw capacity.
- Headroom. Size for 120% of calculated need. Scene activity increases, cameras get added, and a full array silently overwrites the footage you needed.
- Surveillance-rated drives. Desktop drives are not designed for continuous write workloads and fail early in this duty cycle.
The security problem nobody plans for
IP cameras are among the most frequently compromised device classes on the internet. The reasons are consistent and entirely avoidable:
- Default credentials left unchanged on every camera
- The recorder port-forwarded directly to the internet for phone viewing
- Firmware never updated after commissioning
- Cameras placed on the same flat network as business systems
A compromised camera network is not just a privacy problem. It is a foothold inside your network, and it is often the easiest one available.
The fixes are straightforward: put cameras on a dedicated VLAN with no route to business systems, change every default credential and record them properly, provide remote viewing over VPN rather than port forwarding, and include firmware currency in your maintenance schedule.
A workable planning sequence
- Walk the site and list every point where an incident would need to be proved
- Assign each of those points a required level: detection, recognition or identification
- Derive resolution, lens and mounting position from that level and the viewing distance
- Add general coverage cameras for situational awareness
- Confirm the retention period you need — legally, contractually and practically
- Calculate storage from actual configuration, then add 20% headroom
- Design the camera network as a segmented VLAN from the start
- Walk the site again at night before finalising
Done in that order, most sites end up with fewer cameras than they first assumed, positioned considerably better — and footage that is actually usable when it matters.