DORI and Pixel Density: How Far Can a Security Camera Identify a Person?

The four levels#

Level	Pixel density	Approx. pixels across a face	What the operator can do
Detect	25 px/m	4	See that a person is present
Observe	63 px/m	10	See characteristic details (clothing)
Recognise	125 px/m	20	Tell whether it is someone already known
Identify	250 px/m	40	Establish identity beyond reasonable doubt

These densities come from IEC 62676-4 as summarised in Axis Communications' white paper. Pixel density is horizontal pixels on target per metre of scene width at the target distance.

The formula#

A camera with $N$ horizontal pixels and horizontal field of view $θ$ images a scene of width $W = 2 d tan (θ /2)$ at distance $d$ . Pixel density is therefore

ρ_{p x} = \frac{N}{2 d tan ( θ /2 )} \Rightarrow d_{m a x} = \frac{N}{2 ρ _{req} tan ( θ /2 )}

Worked example: a 1080p camera ( $N = 1920$ ) with a 30° lens, required to identify ( $ρ_{req} = 250$ ): $d_{m a x} = 1920/ (2 \times 250 \times tan 1 5^{\circ}) \approx 14.3 m$ .

Tables#

Camera	Detect (25)	Observe (63)	Recognise (125)	Identify (250)
1080p · 90° HFOV	38 m	15 m	7.7 m	3.8 m
1080p · 30° HFOV	143 m	57 m	29 m	14 m
4K · 90° HFOV	77 m	30 m	15 m	7.7 m
4K · 30° HFOV	287 m	114 m	57 m	29 m
1080p · 10° HFOV	439 m	174 m	88 m	44 m
1080p · 3° HFOV	1466 m	582 m	293 m	147 m

Doubling resolution doubles every distance; halving the field of view roughly doubles it again. This is why perimeter cameras that must identify at range use narrow lenses or PTZ zoom, and why a single wide-angle camera on a tall mast can detect across a site but identify almost nothing.

Johnson's criteria#

DORI descends from older work on visual discrimination. In 1958, John Johnson of the U.S. Army night-vision laboratory presented “Analysis of Image Forming Systems,” relating the number of resolvable cycles across a target's critical dimension to the task: roughly 0.5–1 cycle to detect, 3–4 to recognise and 6–8 to identify. One cycle is two pixels at best, so Johnson's numbers translate into pixel requirements once you fix the critical dimension (about 0.5–0.75 m for a person's width, or the face for identification). Thermal imagers are still specified this way.

Why this matters for elevated sensors#

Height improves line of sight (see our viewshed study) but adds distance. A camera on a 40 m mast looking at a point 300 m away is about 303 m from it — nearly no penalty. A camera on a sentry that relocates to within 150 m of a detection, by contrast, halves the distance and doubles the pixel density, which can move a track from “detect” to “recognise.” Distance you can change is worth more than resolution you can buy.

Practical checks#

Use the horizontal pixel count of the stream you actually record, not the sensor's marketing resolution.
Compression, motion blur, low light and IR illumination all reduce effective density; add margin.
For thermal cameras, specify in Johnson-style cycles or in detection/recognition/identification ranges measured for a defined target.
Check DORI at the far edge of the area of interest, not at its centre.

Frequently asked questions#

How do you calculate DORI distance?

d = N / (2 × required px/m × tan(HFOV/2)), where N is horizontal pixels and HFOV the horizontal field of view.

What pixel density is needed to identify a person?

250 pixels per metre under IEC 62676-4 (about 40 pixels across a face).

What are Johnson's criteria?

Rules from 1958 relating cycles across a target's critical dimension to detection (~0.5–1), recognition (3–4) and identification (6–8).

References#

IEC 62676-4:2014, Video surveillance systems for use in security applications — Part 4: Application guidelines. DORI pixel densities as summarised in Axis Communications, “Pixel density based on IEC 62676-4:2014.” whitepapers.axis.com
J. Johnson, “Analysis of Image Forming Systems,” Image Intensifier Symposium, U.S. Army Engineer Research and Development Laboratories, Fort Belvoir, October 1958.

DORI and pixel density: how far can a camera identify a person?