IEC 62676-4:2014 was withdrawn on 9 October 2025 and replaced by Edition 2.0, which turns the six-level ladder the industry still calls DORI into seven levels. 250 px/m, the old Identify figure, is now Characterize; high-certainty identification (Scrutinize) needs 1500 px/m. If your specification still cites the 2014 edition, it cites a withdrawn document.
Four ideas to carry into the next specification conversation you have.
IEC 62676-4 Edition 2.0 (2025): Overview 20, Outline 40, Discern 80, Perceive 125, Characterize 250, Validate 500, Scrutinize 1500 px/m. The 2014 standard never contained the word DORI or the unit px/m in the first place; both are industry shorthand.
It is, and substituting the lens gives the form that ends the megapixel argument: px/m = f / (d × pixel pitch). The pixel count cancels out. Two 8 MP cameras on the same 4 mm lens differ by 38% in pixel density, because one has 1.45 µm pixels and the other 2.0 µm.
True, and there is a ten-second test for whether your own trigonometry is even valid on that lens: divide tan(H/2) by tan(V/2). On a 16:9 sensor a rectilinear lens gives 1.78. A 93°/50° dome gives 2.26, because it is barrel distorted, and its published range is 33% further than the formula predicts. Use the manufacturer's table for wide lenses.
And there is a bigger multiplier nobody applies: blur = speed × exposure × density. Pixel density multiplies the smear, so raising it does not sharpen a moving face. At 250 px/m a walking person at 1/30 s smears 11.7 pixels across a face that is only 40 pixels wide.
Per IEC 62676-4 Edition 2.0 (2025), on the standard 1.6 m target. Ranges are computed for one 8 MP camera at 90° HFOV, centre of frame.
| Level | Min PPM · subject height in px | Use case |
|---|---|---|
| Overview | 20 px/m · 32 px | Presence and count only. One 8 MP camera at 90° reaches 96 m |
| Outline | 40 px/m · 64 px | Body shape and direction of travel. Reaches 48 m |
| Discern | 80 px/m · 128 px | Tell one person from another in the scene. Reaches 24 m |
| Perceive | 125 px/m · 200 px | Clothing, carried objects, activity. Reaches 15 m |
| Characterize | 250 px/m · 400 px | Person type, gait, vehicle brand. This is what the old ladder called Identify. Reaches 7.7 m |
| Validate | 500 px/m · 800 px | Confirm a known identity. Reaches 3.8 m |
| Scrutinize | 1500 px/m · 2400 px | Establish who someone is, with high certainty. Reaches 1.3 m |
Five facts the camera salesperson didn't mention.
Correct, and Saudi Arabia has legislated the point: the MOI conditions document does not tell you to buy 4K cameras, it names the field-of-view class each camera position must achieve and then limits the aiming. Identification cameras within 30° right or left of straight-on and 15° above or below the horizon line, with WDR of at least 110 dB; strong identification (the document's Inspection class, التفتيش) 10° both ways with the image centre between the eyes; plate cameras 30° (Ed. 1 v1.1, pp. 20 to 22). The 15° line alone keeps a camera mounted at 3.5 m at least 7.1 m from the face. Article 12 of the camera law fines per camera: SAR 500 for equipment that misses the document's technical specifications (12a) and SAR 1,000 for a camera not installed as the document sets (12b); a review committee decides which applies (Article 11). The IEC text only says cameras should be mounted around head height.
Size it from the rule the Saudi document actually writes: 1080p minimum, and for motion-linked recording at least 70% of the day at 10 fps, retention not less than 31 days. At 4 Mbps that is 0.94 TB per camera before the capacity clause. Item 10 (p. 13) then asks for “an additional 20% of the total capacity of the recording system”. The 20% is of the total, so the recording may fill 80% of it: need ÷ 0.8, 1.17 TB. That also covers the narrower reading of adding 20% to the need (1.12 TB), so it is the safe design value; if the authority on a project reads it differently, get that in writing. Read the sector tables too: oil and petrochemical sites, power and desalination, public transport stations, and category 4 and 5 hotels and conference venues require 91 days at 15 fps (pp. 38, 40, 45, 75). ATM cameras get 91 days in the shopping-mall table (p. 48), while the banks table (p. 50) prints 19 days and 51 fps for “MTA” cameras, which reads as a transposed 91, 15 and ATM: confirm with the ministry. At 91 days the same arithmetic gives 3.44 TB per camera. These are illustrative estimates at exactly 70% duty: ≥70% is a floor, so a camera that records more needs more, and continuous recording needs about 1.67 TB for 31 days and 4.91 TB for 91. TB here is decimal (1012 bytes), before file-system, RAID and recorder overhead.
Saudi Arabia's grid runs at 60 Hz (Grid Code 2.4.2.1), so mains-driven light ripples at 120 Hz. In an ideal sinusoidal model, a rolling shutter exposure that holds whole cycles, 1/120 s or its multiples, cancels banding; PWM dimming, mains drift between 59.9 and 60.1 Hz and camera timing can leave some even then. The European 1/100 s keeps 16% of the ripple as banding. But the shutter that freezes a walk at Characterize density is about 1/350 s, and that keeps 82%. So the banding in the frame is the driver's modulation times a share the shutter sets: an unfiltered driver bands at 82%, a driver inside IEEE 1789's no-effect line at 120 Hz (Mod% 4.0) at 3.3%. IEEE 1789 rates risk to people, not camera compatibility, so that 3.3% is a model result, not a CCTV threshold; test banding on site with the real driver and camera. Sensor makers say the same from the other side: below one light period, flicker from AC-driven lamps cannot be avoided. Write the driver's Mod% into the lighting specification, because IEEE 1789 itself has been inactive since March 2026.
Each Saudi class is defined first as the target's image size as a share of the view on the display, on a 1.6 m standard target: Identification not less than 120%, Recognition 50%, Observation 25%, Detection 10%, Monitoring 5%. Pixel density may be substituted only in exceptional cases and with prior Ministry approval, so a px/m-only specification is a non-compliant one. The percentages match IEC 62676-4:2014 Table 3, though not one column: 120% is its NTSC value, where PAL gave 100%, while 50% and 25% are the PAL values. The px/m substitutes (1000, 250, 125, 62, 25 and 12.5 from strong identification down to monitoring) are the 2014 values, except that Observation is printed as 62, and plate reading adds a seventh rung at 400 px/m (Ed. 1 v1.1, pp. 20 to 22). Signage is not a PDPL requirement either: it comes from the camera law, and PDPL then governs what happens to the data afterwards.
The conditions document carries a mandatory section titled Security Lighting: the camera field of view fully lit with no dark areas, and, at external sites, infrared allowed in secondary areas or wherever visible light is not enough (p. 25), never reflecting into the camera, and no harm to the safety of others, such as drivers' vision. It gives no lux, ratio, colour or flicker figure, so the designer brings the numbers. IEC 62676-4 asks for a max-to-min ratio ideally 10:1 or better, light directed along the expected direction of travel so faces are lit, and luminaires preferably above the camera and at least 2 m from it, because they attract insects. The UK Home Office's 2009 manual recommends no more than 3:1. On 6 m poles of an idealised luminaire, that is the difference between 31 m and 18 m spacing. On WDR itself, note the trade: turning it on typically halves the frame rate and doubles the exposure window, which blurs the face you bought it to see.
Lighting calculations report horizontal illuminance on the ground. A camera needs vertical light on the face, from its own side. Along a row of poles the lux plot peaks directly under each pole, which is exactly where a face looking along the row gets 14% of that reading; about 3 m before the pole it gets 78%. So a ground-level horizontal illuminance plot does not establish facial vertical illuminance. Road lighting addresses faces through the semi-cylindrical classes of EN 13201-2, written for the human eye; we have not reviewed that standard in full, so read its class values there before citing them. No CCTV document reviewed here sets a facial vertical illuminance, which is why a specification has to name one and have it measured.
In the withdrawn 2014 edition, clause 6.7 gives every level a design figure, Identify included: 100% of screen height on PAL, 120% on NTSC, or 4 mm per pixel, which is 250 px/m. The acceptance test in clause 13.3.2 gives millimetres for inspect, recognise, observe, detect and monitor, and for identification sends you to Annex B: nine printed faces, a height stick and an operator, at a screen height the Operational Requirement sets. The annex passes at 4 out of 6, reached by naming one face and getting only the broad category right on the other, while the number-plate test beside it needs every character. The 2025 edition's contents list a normative Annex B test protocol, but we could not access its procedure or scoring, so the contract should state the acceptance method and pass criteria itself. On Saudi projects the MOI document adds its own image acceptance (p. 27): a clear, stable image in moderate wind, obstructions under 5% of the frame and none at its centre. Pixel density is a design input, not a guarantee, and the IEC levels assume a human viewer; face or plate analytics follow the vendor's own requirements.
NLC does not sell cameras. It sells the thing that makes them work: illuminance in the field of view that meets the conditions document's Security Lighting clause, a max-to-min ratio at or better than the 10:1 IEC 62676-4 recommends, vertical light on faces from the side the cameras look from, and drivers quiet enough at 120 Hz for a shutter fast enough to freeze a walk, without harming drivers' vision on the road beside it.
Practical lighting know-how from NLC Academy. No spam, unsubscribe anytime.
By subscribing you agree to receive NLC emails.
The standards and regulations behind every number on this page.
A 20-slide explainer, ready to post straight to LinkedIn as an image carousel.
