CCTV Lens Selection Technical Guide: Focal Length, Pixel Density & the DORI Standard in Practice

TL;DR

• Face identification depends on pixel density (px/m), not megapixels — EN 62676-4 DORI: 250 px/m to identify, 125 px/m to recognize.
• Core formulas: FOV width = 2 x Distance x tan(HFOV/2); pixel density = horizontal resolution / FOV width.
• 2.8mm suits entrances within 5m, 4mm is the general-purpose choice, 6mm+ for corridors and perimeters — wide lenses cannot identify faces beyond ~8m.
• Night vision is decided by sensor size (1/1.8" beats 1/2.8") and aperture (F1.0 admits 4x the light of F2.0), not by IR LED count.

1. Why 4MP Still Can't Read a Face

Total pixels are spread across the entire scene, while a face occupies only a tiny fraction. A 4MP camera on a 2.8mm wide lens covering a parking lot may deliver under 20 px/m on a person — barely "that is a human". The same camera with a 6mm lens on a corridor exceeds 250 px/m, reaching identification level. The industry measures this with the DORI system defined in EN 62676-4 (identical to IEC 62676-4):

2. The Core Math: Three Parameters, One Formula Chain

Lens selection needs only three parameters: mounting distance D, horizontal field of view HFOV (determined by focal length and sensor size), and horizontal resolution. Two steps:

Step 1: FOV width W = 2 x D x tan(HFOV / 2)

Step 2: Pixel density PPM = horizontal resolution / W

(D = distance to target in meters; HFOV from the datasheet; horizontal resolution: 1080p=1920, 4MP=2560, 4K=3840)

2.1 Worked Example: Why a Wide Lens Fails at the Gate

Scenario: gate 6m wide, camera mounted 10m away, 2.8mm lens (1/2.8" sensor, HFOV ~100 deg), 1080p.

W = 2 x 10 x tan(50 deg) ~ 23.8 m

PPM = 1920 / 23.8 ~ 81 px/m

Result: 81 px/m is only O-level — you see what happened, but not who did it.

Same scene with 4mm (HFOV ~78 deg): W ~16.2m, PPM ~118 px/m — near R-level. With 6mm (~50 deg) + 4MP (2560px): W ~9.3m, PPM ~275 px/m — I-level, strangers identifiable. Same distance, same position — just the right focal length and resolution.

2.2 Reverse Engineering: Define the Need First

Work backwards: set target distance, set the DORI level, compute the maximum FOV width, then pick the focal length. To identify faces at 15m (I-level 250 px/m, 4MP): W <= 2560/250 = 10.2m, so 2x15xtan(theta/2) <= 10.2 gives theta <= 37.6 deg — choose 8mm or longer.

3. Focal Length Lookup Table (typical 1/2.8" sensor, 4MP)

Reference values for 4MP (2560x1440); always confirm with the specific model datasheet — HFOV varies up to +/-10% between vendors at the same focal length:

4. Sensor Size & Aperture: What Really Decides Night Vision

Night image quality is decided by light intake: sensor area and aperture F-number — IR LED counts and "XXm night vision" claims are secondary.

4.1 Sensor Size: Bigger Is Simply Better

Datasheet trap: two cameras both labeled "4MP" — one 1/2.8", one 1/1.8" — have pixel sizes differing by almost 2x and are completely different machines at night. For night-critical points (gates, perimeters), sensor size outranks resolution.

4.2 Aperture: Smaller F-Number, More Light

Light intake is inverse to the square of the F-number: F1.0 admits 4x the light of F2.0; F1.6 admits ~56% more than F2.0. "Full-color night vision" is essentially large sensor + F1.0-F1.2 aperture + noise reduction. Within the same series, pick the smaller F-number version — typically 10-20% more cost for a qualitative night-vision jump.

5. Fixed vs Varifocal vs PTZ

6. The Five Most Common Selection Mistakes

1. Wide-angle for a gate panorama AND face ID: panorama and identification conflict — use a wide overview camera plus a telephoto on the choke point, not one camera for both
2. Blindly going 4K: 4K spread over a wide FOV still lacks pixel density, while costing 2.2x storage/bandwidth. Use 4K on telephoto for detail in large scenes, not as a substitute for more points
3. Wide lens in corridors: with a 20m-deep corridor, 90% of a 2.8mm frame is walls. Use a telephoto down the corridor or a corridor-mode (9:16) model
4. Ignoring mounting height: at 6m height you record the top of heads. Face points belong at 2.5-3.5m with <=15 deg tilt — higher is not better
5. Not testing WDR at backlit doors: bright outside, dark inside — faces go black without WDR. Specify >=120dB for entrances and field-test with a strong flashlight, not the brochure

7. FAQ

Q1: 2.8mm or 4mm — how do I choose?

By target distance: under 5m choose 2.8mm, 5-10m choose 4mm, beyond 10m at least 6mm. Alternative test: do you need "what happened in this area" (2.8/4mm) or "who is this person" (use the lookup table by distance)? If unsure, choose a motorized varifocal model and adjust remotely after installation.

Q2: Is higher resolution always better? What does 8MP add over 4MP?

Same focal length and sensor size: 8MP (4K) gives 1.5x linear pixel density and ~50% more identification distance. Costs: ~2.2x storage and bandwidth, less light per pixel (night vision worsens at equal sensor size), 30-50% higher price. Rule: daytime large-scene distance identification -> 4K; night-first -> larger-sensor 4MP; tight budget -> spend on more points, not more pixels per point.

Q3: Does night vision get worse at the telephoto end of a varifocal lens?

Yes — physics. Most motorized varifocal lenses have variable aperture: F1.6 wide, closing to F2.8 or smaller at tele — a 3x light difference. Check whether the datasheet says "F1.6 (W) - F2.8 (T)" or constant aperture. For night-critical points with varifocal cameras, evaluate night vision at the telephoto aperture.

Q4: How do I verify a claimed "50m identification distance"?

Ask for the calculation, or compute it yourself: identification distance = horizontal resolution / 250 px/m / (2 x tan(HFOV/2)). For 4MP at HFOV 38 deg (8mm): 2560/250 = 10.24m FOV width, distance ~15m. Any claim far from the formula deserves a question about test conditions. Harder proof: demand a PoC — mount a sample at the real point and record actual night footage before ordering.

Q5: What is special about license plate capture points?

Plates need ~500 px/m (double the face requirement) and are shutter-sensitive. At 30 km/h, shutter must be <=1/500s to freeze the plate — plate points need strong supplementary lighting to support fast shutter. For professional use, choose a dedicated ANPR camera (built-in illumination and plate algorithms); tuning a general camera is only a stopgap.