Projects that open with the question “How many cameras do we need?” usually end in one of two outcomes: either more cameras than necessary are bought, or on the day of an incident the recording doesn’t show a face, plate or pallet clearly enough. Both have the same cause. The number of cameras was chosen from a catalogue before the goal was defined. Below is the sequence to follow when designing an IP camera system for a corporate facility, from defining the goal to the specification and acceptance criteria.
1. Put the goal and scenarios in writing
System design starts with a scenario matrix. For each critical area, three questions are answered: What incident could happen here? What footage is needed when it happens? Who will watch or review the footage, and when? If the goal at a shipping dock is “to show whether pallet damage occurred during loading”, the camera must see the back door of the vehicle and the pallet at recognition level, not the whole dock. The overview is another camera’s job.
The matrix is prepared together with the security, OHS, production, logistics and IT teams. Its output is a table with columns for area, purpose, required image level, recording mode (continuous/event-based) and access permissions. A common gap is having only the security team fill in this table. When needs such as reviewing production stoppages or analyzing OHS accidents emerge after installation, camera positions change and cable is pulled a second time.
The matrix also puts what’s out of scope in writing. Areas deliberately not monitored (privacy zones such as changing rooms and rest areas) and parts of the view to be masked are decided at this stage, so there’s no debate later.
2. Define the image level by pixel density
In the IEC 62676-4:2014 table, image goals are expressed as the pixel density the target occupies in the image. The widely used classification is DORI:
| Level | Purpose | Approximate pixel density |
|---|---|---|
| Detection | Noticing that a person is present | ~25 px/m |
| Observation | Following movement and behavior | ~62 px/m |
| Recognition | Recognizing a familiar person | ~125 px/m |
| Identification | Design goal for identity detail: aiming for detail in the captured image that helps assess identity | ~250 px/m |
About 250 px/m is the reference in the IEC 62676-4:2014 table. In testing, target distance, face angle, lighting, motion blur and compression are assessed together. IEC 62676-4:2025, published in October 2025, redefined pixel densities; it uses seven categories and new test charts instead of four DORI levels. The table above belongs to the 2014 edition, not the 2025 edition. In a new project, state which edition and which test method the specification is based on.
The calculation is simple: pixel density = horizontal resolution ÷ field-of-view width at the target distance (m). If a camera with 2560 pixels of horizontal resolution (4 MP) sees an area 10 meters wide, the density is 256 px/m; on paper it meets the identification reference in the 2014 table (about 250 px/m). If the same camera sees a 30-meter area, it drops to 85 px/m and only meets the observation level. DORI distances in manufacturer brochures assume ideal light and contrast. Night noise and compression reduce real detail.
Mounting height and vertical angle also count. If the camera looks straight down from 5–6 meters, even with enough horizontal density the recording shows the top of a head and a cap peak instead of a face. At door and turnstile points where identification is needed, placing the camera closer to and lower than the target often gives better results than a longer lens. If the angle is wrong, resolution won’t save it.
3. Survey the site by day and at night
The survey records mounting surfaces, heights, existing lighting, backlight sources (sun, vehicle headlights, glass façades), dust and humidity, cable routes and cabinet locations. Decisions made for outdoor and door points without a night visit are a frequent source of errors: IR reflection, poor lighting and headlight glare only show up at night. Taking a few minutes of sample recording with a test camera at critical points is the cheapest verification.
For each camera point, measurements and photos are kept together: mounting height, distance to target, expected field-of-view width and the real cable distance to the nearest cabinet. Cable distance isn’t measured as the crow flies on the plan; tray, vertical runs and slack inside the cabinet are added. A copper channel is limited to 100 meters, and patch cords count toward that limit. Points beyond this distance need fiber or a field switch, and that should come out in the survey. Not on installation day.
4. Choose the camera type and lens
- Fixed cameras produce evidence; they’re the basic choice at critical entrances, checkouts and docks.
- Varifocal lenses let the field of view be adjusted to the site during installation.
- PTZ is used for detail and tracking in large areas; since it looks in one direction at a time, it doesn’t replace fixed cameras (PTZ design).
- Panoramic/fisheye cameras provide general situational awareness; detail is low at the edges.
- Thermal cameras are for detection in darkness and on long perimeters, not for identification (thermal camera guide).
Selection weighs sensor size, WDR performance, night mode, IR range, IP/IK protection rating, operating temperature, supported codecs, ONVIF profiles and the manufacturer’s security update policy together. Choosing by a single parameter, usually megapixels, is the most common mistake in the field: at a backlit entrance, an 8 MP camera with weak WDR gives less usable footage than a 4 MP camera with good WDR.
Another common expectation is to capture both the whole hall and the face of someone walking in the door with a single wide-angle camera. The pixel density calculation shows from the start that this isn’t possible. The overview and the identification point are separate cameras.
5. Calculate network, PoE and power capacity
The worst-case power draw of each camera (IR, heater, PTZ motor) is added up and compared with the switch’s total PoE budget; details are in the PoE power budget guide. For bandwidth, camera bitrates are added up and the uplink capacity from the access switch to the recording server is checked. Cameras are planned in a separate segment (camera VLAN). UPS runtime for switches and recorders is calculated according to the organization’s outage scenario.
It isn’t only the recording stream that goes into the uplink calculation. If operators watch many cameras from the main stream at once, a second load is added on top of recording traffic; using the VMS’s sub-stream support for mosaic viewing noticeably reduces this load. On the switch side, IGMP snooping, PoE priority and monitoring (SNMP, syslog) are written into the specification. In a system built with unmanaged switches, nobody knows when a port goes down; missing recordings are noticed only after an incident.
6. Define the recording, storage and management architecture
The number of cameras and sites, the need for mixed brands and permission expectations decide between an NVR and an enterprise VMS (VMS or NVR?). Storage is calculated from bitrate, recording mode and retention period (How many days does an NVR record?). The retention period itself is an organizational and legal decision, not a technical one; keeping recordings for the necessary period and destroying expired ones must be planned.
Storage is calculated with the night-time bitrate. Under IR and in low light, the same camera can produce several times its daytime data; a disk sized from a daytime measurement can’t meet the target retention. The disk failure scenario should also be written into the project: RAID level, hot spare, who receives the failure notification and within how many days a failed disk is replaced. A RAID whose alerts nobody reads runs silently until the second disk fails.
7. Write the specification and acceptance criteria together
A brand-neutral specification should include performance goals such as “recognition level at this point at night” instead of minimum resolution and sensor features, ONVIF profile requirements, cybersecurity requirements (HTTPS, user roles, update support), PoE class and documentation deliverables. Acceptance criteria belong in the same document:
- Playback review by day and night for each camera, checking the target level
- Measuring recording continuity and actual retention
- Clock synchronization (NTP) and consistent event timestamps
- A security check confirming default passwords are removed, unnecessary services are disabled and firmware versions are recorded
- Delivery of the as-built plan, device inventory (IP, port, model, firmware) and the user/permission matrix
Don’t leave acceptance testing to a form the installation team fills in and signs itself. Playback review is done with a person responsible from the operations side; for each camera, the result “target level met / not met” is recorded with a screenshot. Looking at the live image isn’t enough: acceptance is done on the compressed recording, because that’s what you’ll have on the day of an incident.
8. Common system design mistakes
- Expecting both an overview and identification from a single wide-angle camera
- Not accounting for backlight and night conditions
- Calculating the PoE budget with nominal consumption and forgetting heater and IR load
- Calculating recording capacity with the daytime bitrate
- Leaving cameras on the same segment as the corporate network
- Measuring cable distance as a straight line on the plan without adding tray and vertical runs
- Commissioning without configuring NTP, so at the moment of an incident the recording times of different cameras don’t match
- Trying to set acceptance criteria at the end of the project
Frequently asked questions
Is a floor plan required for system design?
A scaled plan speeds up the process a lot and lets the layout be shown on the plan. If there’s no plan, a simple layout drawing can be prepared from measurements taken during the survey.
Can you tell us the number of cameras up front?
The number of cameras is the result of the design. It becomes clear once the critical points and the image level needed at each point are defined; in the initial meeting only a rough range can be given.
Should the specification name a brand?
In corporate projects, writing performance and compliance criteria lets you get comparable bids from more than one manufacturer. Mandatory cases such as compatibility with an existing VMS are stated separately.
The output of system design lets you get comparable bids from different contractors and accept the installation against measurable criteria. For your facility, see our IP camera system design service, or if you’re still at the decision stage, our survey and consulting service.