Design Right AI · Reference

Field of view and DORI zones

What DORI actually specifies, how to calculate the distance at which a camera can identify rather than merely detect, and why adding megapixels to a wide lens does not buy you the recognition range you think it does.

DORI is a pixel-density standard, not a distance rating

EN 62676-4 defines four operational levels by pixels per metre of scene — how much sensor resolution lands on a target at a given distance. The distances that get quoted on a spec sheet are derived; the px/m figures are the standard.

Identify250 px/m Enough detail to identify a specific individual beyond reasonable doubt.

Recognize125 px/m Determine with high certainty whether a person is someone previously seen.

Observe62 px/m Characteristic details visible — clothing, general appearance, behavior.

Detect25 px/m Determine with high certainty that a person is present. Not who.

The gap between Detect and Identify is a factor of ten. A camera that "sees" the whole car park almost certainly cannot identify anyone in it — and that distinction is the one that matters when footage is asked for after an incident.

The calculation

hFOV = 2 × atan( sensor width ÷ (2 × focal length) )

distance for level L = horizontal pixels ÷ (2 × tan(hFOV ÷ 2)) ÷ L

Worked for a 4 MP camera (2688 px wide) on a 1/2.8" sensor (5.18 mm) with a 4 mm lens: hFOV is 65.8°, so the scene width at distance d is 1.29d, and pixel density is 2080÷d px/m. Setting that equal to each DORI threshold gives 8.3 m to identify, 16.6 m to recognize, 33.2 m to observe, 83.0 m to detect.

DORI distance by focal length

4 MP (2688 px) on a 1/2.8" sensor. Longer lens narrows the field and pushes every zone further out — you are not gaining resolution, you are spending it over less width.
Lens	hFOV	Identify	Recognize	Observe	Detect
2.8 mm	85.5°	5.8 m	11.6 m	23.2 m	58.1 m
4 mm	65.8°	8.3 m	16.6 m	33.2 m	83.0 m
6 mm	46.7°	12.5 m	24.9 m	49.8 m	124.5 m
12 mm	24.4°	24.9 m	49.8 m	99.6 m	249.1 m

Resolution buys less range than people expect

All on the same 4 mm lens. Doubling horizontal pixels only extends each zone by the same factor — going 2 MP to 4K roughly doubles identify range, from 5.9 m to 11.9 m.
Resolution	Horizontal px	Identify	Recognize	Observe	Detect
2 MP / 1080p	1920	5.9 m	11.9 m	23.7 m	59.3 m
4 MP	2688	8.3 m	16.6 m	33.2 m	83.0 m
4K / 8 MP	3840	11.9 m	23.7 m	47.4 m	118.6 m

The megapixel trap

A 4K camera on a 2.8 mm lens identifies to about 8.3 m. A 4 MP camera on a 4 mm lens identifies to the same 8.3 m — at half the storage and half the bitrate. Resolution spread across a wider field buys nothing. Choose the lens for the range you need, then the resolution. Doing it the other way round is how sites end up with 4K cameras that still cannot identify anyone.

Typical field of view by camera type

Planning defaults used by the Design Right AI coverage engine when a specific lens has not yet been selected.
Camera type	Horizontal FOV	Vertical FOV	Typical use
Dome	100°	75°	General interior coverage
Turret	95°	70°	Interior and sheltered exterior
Bullet	90°	50°	Directed exterior runs
PTZ (at typical zoom)	60°	40°	Wide areas with operator control
Thermal	50°	38°	Perimeter detection
LPR	35°	26°	Plate capture on a defined lane
Fisheye	180°	180°	Whole-room overview
Multi-sensor	180°	90°	Corner and intersection coverage

Thermal does not use DORI

Thermal cameras are rated against the Johnson criteria, which counts pixels across the target's critical dimension rather than pixels per metre of scene. The thresholds are much lower, because a thermal signature is a silhouette against background temperature, not a face.

Detection — roughly 1.5 pixels across the target
Recognition — roughly 6 pixels across the target
Identification — roughly 12 pixels across the target

This is why a modest thermal sensor detects a walking person at several hundred metres while a far higher-resolution visible camera cannot. It is also why thermal "identification" range must never be read as identifying a person in the evidentiary sense — it means classifying the target as human rather than animal or vehicle.

What DORI does not tell you

Nothing about mounting height or tilt. DORI is a horizontal density figure. A camera at 3.5 m with 15° of tilt has a blind spot beneath it several metres deep, and its useful band starts past that.
Nothing about light. The px/m thresholds assume the target is adequately lit. At night, with IR at range or a wide-open aperture, effective detail can fall well below what the geometry predicts.
Nothing about the target's orientation. Identify assumes a face toward the lens. The same 250 px/m on the back of a head identifies nobody.
Nothing about compression. Pixels the encoder threw away are not available to an investigator, regardless of what the sensor captured.

Common questions

Is DORI the same as PPF?

Same idea, different unit. PPF is pixels per foot, common in North America; DORI is specified in pixels per metre. 250 px/m is roughly 76 px/ft. Mixing the two is a frequent source of designs that miss their target by a factor of three.

Which zone should I design to?

Work backwards from the purpose. Access points, tills and any location where the question will be "who was that" need Identify. Corridors and general movement need Recognize or Observe. Perimeter and open ground where the question is "was anyone there" need Detect. Designing everything to Identify is how budgets get destroyed.

Does a varifocal lens change the answer?

Yes — every zone moves as you zoom. A 2.8–12 mm varifocal set at 2.8 mm identifies to about 5.8 m; the same camera at 12 mm identifies to 24.9 m over a much narrower field. Commission the lens at the setting the design assumed, and record it.

How does this affect a PTZ?

A PTZ's DORI figures apply only at the zoom it happens to be at. Its parked position is the only one you can rely on for evidentiary coverage, which is why PTZs supplement fixed cameras rather than replacing them.

Designing coverage on a real floor plan?

Design Right AI draws DORI zones directly on the plan as you place each camera — lens, mounting height and tilt applied to the actual geometry — so the coverage a customer is shown is the coverage the bill of materials will deliver.