Shenzhen Link-Card Tech Co., LTDMenu

Infrared Lens Parameters Explained: Focal Length, F-Number, Field of View, and Detection Range

Tech Blog · Infrared Optics

Infrared Lens Parameters Explained: Focal Length, F-Number, Field of View, and Detection Range

Published 2026-09-05 · Link-Card Technology

Focal length, F-number, and pixel pitch are the three fundamental quantities of an infrared lens: focal length and sensor format jointly set the field of view, focal length and pixel pitch set the observation range, and the F-number sets throughput and the diffraction limit. Working from 17/12/8 μm pixel pitches and the two mainstream 640×512 and 384×288 formats, this article derives the parameters for eight focal lengths from 9.1 to 100 mm, presents the full matrices, and estimates the solar image size on the detector — the geometric basis of sun-burn protection algorithms.

1. Three fundamental quantities and six sensor formats

The specification of an infrared lens reduces to three fundamental quantities. The focal length f sets the angular magnification of the system and is the first variable behind "how wide and how far"; it usually appears directly in the product name (9.1 mm, 19 mm, 100 mm). The F-number, defined as focal length divided by entrance-pupil diameter (F/# = f/D), does not affect the field of view; it governs two other things — the irradiance reaching the focal plane and the smallest angular detail diffraction allows. The pixel pitch p (17 μm, 12 μm, 8 μm) is the sampling interval of the detector and, together with focal length, determines the solid angle covered by a single pixel.

These combine into the two most frequently used performance quantities. The field of view is fixed jointly by focal length and sensor width: HFOV = 2·arctan(W/2f), where W = N×p is the physical width of the detector. The instantaneous field of view (IFOV) = p/f is the angular size of one pixel, in mrad, and is the conversion factor from observation range to pixel count. Both formulas run through this article, and every number can be recomputed from them.

First, the geometry of the six baseline formats. 384×288 and 640×512 are the two mainstream resolution classes of today's uncooled cores; 17 μm is the previous-generation mainstream process, 12 μm is the current shipping workhorse, and 8 μm (including 8.5 μm) is the emerging generation that has been entering volume production in recent years.

Table 1 — Geometry of the six baseline sensor formats
Format (resolution × pitch)Width × height (mm)Diagonal (mm)Nyquist frequency (lp/mm)
640×512 @ 17 μm10.88 × 8.7013.9329.4
640×512 @ 12 μm7.68 × 6.149.8441.7
640×512 @ 8 μm5.12 × 4.106.5662.5
384×288 @ 17 μm6.53 × 4.908.1629.4
384×288 @ 12 μm4.61 × 3.465.7641.7
384×288 @ 8 μm3.07 × 2.303.8462.5

The Nyquist frequency fN = 1/2p is the upper spatial frequency the lens must resolve: the smaller the pixel, the higher the MTF demand on the lens, developed further in Section 4. Table 1 also implies a fact that is easy to overlook: field of view is not a property of the lens alone but of the "lens + sensor" combination. The same 19 mm lens delivers a 32.0° HFOV on a 640×512@17 μm array and only 9.2° on a 384×288@8 μm array. Any field-of-view discussion must begin by fixing the sensor format.

2. The field-of-view matrix: eight focal lengths × two resolutions × three pitches

Substituting into HFOV = 2·arctan(W/2f) produces the matrices below. Example — 640×512@12 μm with a 19 mm lens: W = 640×12 μm = 7.68 mm, so HFOV = 2·arctan(7.68/38) = 22.9°; VFOV = 2·arctan(6.14/38) = 18.4°. At small angles the vertical-to-horizontal ratio approaches the sensor aspect ratio (about 0.8 for 640×512, 0.75 for 384×288); the industry convention is to rate lenses by HFOV.

Table 2 — Horizontal field of view on the 640×512 format (°)
Focal lengthHFOV @ 17 μmHFOV @ 12 μmHFOV @ 8 μm
9.1 mm61.745.831.4
13 mm45.432.922.3
19 mm32.022.915.4
25 mm24.617.511.7
35 mm17.712.58.4
50 mm12.48.85.9
75 mm8.35.93.9
100 mm6.24.42.9
Table 3 — Horizontal field of view on the 384×288 format (°)
Focal lengthHFOV @ 17 μmHFOV @ 12 μmHFOV @ 8 μm
9.1 mm39.528.419.2
13 mm28.220.113.5
19 mm19.513.89.2
25 mm14.910.57.0
35 mm10.77.55.0
50 mm7.55.33.5
75 mm5.03.52.4
100 mm3.72.61.8

The computed values can be cross-checked against vendor listings: a Chinese lens catalog entry for an "FL 75 mm F/1.2, 640×512-12 μm" lens states (H) 5.9°, matching the 5.86° computed for 75 mm@12 μm in Table 2. Figure 1 plots the field-of-view curves for the three pitches on 640×512; the 384×288 curves have the same shape, shifted down.

0 10 20 30 40 50 60 0 25 50 75 100 HFOV (°) Focal length f (mm) 640×512 @ 17μm @ 12μm @ 8μm
Figure 1 — HFOV versus focal length on the 640×512 format at 17/12/8 μm pixel pitch, computed from HFOV = 2·arctan(W/2f).

Three regularities follow from the tables:

3. Observation range: from IFOV to the Johnson criteria

Range starts from a definition of “seeing”. The Johnson criteria, at 50% probability, count line pairs across the target’s critical dimension, not pixels: 1.0 for detection, 4.0 for recognition, and 6.4 for identification. At Nyquist, one line pair is two pixels, so the thresholds are 2, 8, and 12.8 pixels. Catalogs that write 1.5 / 6 / 12 pixels skip that sampling step and come out about one third long; they will not match Table 4. The critical dimension used here is the 0.5 m width of a standing person (1.8 m × 0.5 m), the same man target Lynred prints on its datasheets. A 0.75 m shoulder width stretches every range by another 1.5×. At range D the target subtends L/D, one pixel subtends IFOV = p/f, and the pixel count is n = L·f/(D·p), so

D = L·f / (n·p)

Recognition range is one quarter of detection range; identification range is 2/12.8 = 1/6.4 of it. Table 4 is a sampling upper bound for a clear atmosphere and enough thermal contrast. NETD, extinction, and the display chain all shorten it. A catalog kilometre figure is comparable only after its pixel threshold and critical dimension are known.

Table 4 — IFOV and Johnson ranges (0.5 m person width, 50% probability, 2 / 8 / 12.8 pixels, clear atmosphere; range in m)
Focal lengthIFOV (mrad)Detection / recognition / identification (m)
17 μm12 μm8 μm17 μm12 μm8 μm
9.1 mm1.8681.3190.879134 / 33 / 21190 / 47 / 30284 / 71 / 44
13 mm1.3080.9230.615191 / 48 / 30271 / 68 / 42406 / 102 / 63
19 mm0.8950.6320.421279 / 70 / 44396 / 99 / 62594 / 148 / 93
25 mm0.6800.4800.320368 / 92 / 57521 / 130 / 81781 / 195 / 122
35 mm0.4860.3430.229515 / 129 / 80729 / 182 / 1141094 / 273 / 171
50 mm0.3400.2400.160735 / 184 / 1151042 / 260 / 1631563 / 391 / 244
75 mm0.2270.1600.1071103 / 276 / 1721562 / 391 / 2442344 / 586 / 366
100 mm0.1700.1200.0801471 / 368 / 2302083 / 521 / 3263125 / 781 / 488

Four readings of Table 4 deserve emphasis:

The formula runs backwards. Detecting a 0.5 m wide person at 2 km on a 12 μm detector (2 pixels) needs f ≥ 2000×2×12 μm / 0.5 m = 96 mm, the 100 mm class. Recognition at 1 km (8 pixels) needs f ≥ 192 mm, which a 100 mm lens does not provide. A 50 mm lens still labeled “2 km detection” should be checked for a 1.5-pixel threshold, or for a critical dimension of 1.8 m rather than 0.5 m. Any one of those changes the kilometre figure.

4. F-number: throughput, the diffraction limit, and pixel matching

The F-number does not change the field of view; it changes the radiant power per unit image area and the smallest detail diffraction permits. Throughput: image-plane irradiance ∝ τ/(F/#)² (τ is lens transmission). Stopping down from f/1.0 to f/1.2 reduces flux to 69%; to f/1.4, 51%. NETD tracks flux directly, which is why uncooled cores rate NETD at f/1.0 — vendor datasheets read, for example, InfiRay E6 Pro (640×512, 12 μm) NETD ≤ 25 mK @ f/1.0, and Raythink industrial handheld series ≤ 40 mK @ F/#1.0. Always check which F-number a NETD figure was rated at before comparing two cores.

Diffraction: the Airy disk diameter of a circular aperture is 2.44λF/#, i.e. 24.4 μm at f/1.0 and 29.3 μm at f/1.2 for λ = 10 μm. LWIR wavelengths are about 20× those of visible light, so the Airy disk is the same order as the pixel pitch — diffraction is not a negligible detail but a hard constraint in direct conflict with sampling. The criterion is the lens diffraction MTF at the Nyquist frequency: with normalized spatial frequency u = λF/#/2p, the diffraction-limited MTF of a circular pupil is MTF(u) = (2/π)[arccos u − u·√(1−u²)]. Evaluating for every combination gives Table 5.

Table 5 — Diffraction-limited MTF at the Nyquist frequency (λ = 10 μm, circular pupil)
Pixel pitchf/1.0f/1.1f/1.2f/1.4
17 μm0.630.600.560.49
12 μm0.490.440.390.30
8 μm0.260.200.140.05

System MTF at Nyquist is generally required to stay above 0.3, with headroom reserved for the lens as a single element. Table 5 draws the F-number/pitch matching boundaries:

Depth of focus is the other side of the F-number: depth of focus ≈ ±2λ(F/#)², i.e. about ±20 μm at f/1.0, ±29 μm at f/1.2, ±39 μm at f/1.4 for λ = 10 μm. At f/1.0 the focus and thermal-drift budget is only tens of micrometres — one physical reason small-F/# lenses cost more to manufacture and align (athermalization was covered in article 04 of this series).

Lens catalogs corroborate the matching rules: athermalized fixed-focus lenses from 9.1 to 50 mm cluster at f/1.0–1.1, while catalog products at 75 mm and 100 mm commonly run f/1.2 (e.g. Stanford Materials 75 mm f/1.2, a China-made catalog FL 75 mm F/1.2 for 640×512-12 μm, and 100 mm f/1.2 units for 384×288/640×512-12 μm) — at long focal lengths the entrance pupil grows, so stopping down slightly controls diameter and weight while the 12 μm diffraction budget at f/1.2 (0.39) remains acceptable. Cooled cores add one more constraint: the lens F/# must match the cold shield F/# or narcissus and stray-radiation artifacts appear; uncooled cores have no cold shield and are exempt.

5. Solar image estimation: the geometric basis of sun-burn protection

A microbolometer converts absorbed radiant power into membrane temperature rise, and the lens concentrates all power collected by the entrance pupil onto the focal plane. To a ground observer the Sun is an intense point source of about 0.53°; focused by the lens it forms a tiny image of extreme irradiance, and the local temperature rise can permanently damage the membrane and bridge circuitry — vendor manuals carry a uniform warning, and FLIR's support page states it plainly: do not point the infrared camera at intensive energy sources, with or without the lens cover. Cameras commonly implement sun-burn protection in software: locate the solar image and mask or replace it. Whatever the detection route, the algorithm rests on a set of geometric parameters — how large the solar image is, how many pixels it covers, and where it lands.

The solar angular diameter varies annually between 0.527° and 0.545° (31.6′–32.7′) with the Earth–Sun distance; computations here take 0.533° (32 arcmin). The geometric image diameter depends on focal length alone:

d = 2·f·tan(0.533°/2) ≈ 9.3·f μm (f in mm)

The pixel coverage depends on the pitch: d/p pixels across the diameter and about (π/4)(d/p)² pixels over the disk. Table 6 lists every case. Two remarks: the solar image size in pixels is independent of resolution — it depends only on f and p, and 640×512 versus 384×288 changes only how much of the frame the solar image occupies; and on a smaller-pitch array the covered pixel count grows as (p₁/p₂)².

Table 6 — Solar image size on the detector (solar angular diameter 0.533°; frame share computed as 0.533°/HFOV at 12 μm)
Focal lengthSolar image diameter (μm)Diameter in pixelsTotal pixels coveredShare of 640×512@12 μm widthShare of 384×288@12 μm width
17 μm12 μm8 μm17 μm12 μm8 μm
9.1 mm855.07.110.61939881.2%1.9%
13 mm1217.110.115.140801791.6%2.7%
19 mm17710.414.722.1851703832.3%3.9%
25 mm23313.719.429.11472956643.1%5.1%
35 mm32619.227.140.728857813014.3%7.1%
50 mm46527.438.858.1588118026556.1%10.1%
75 mm69841.058.187.21323265559749.1%15.1%
100 mm93054.777.5116.3235247201062012.1%20.2%

Do not scale the spot from the full-spectrum 1000 W/m² at the ground. Germanium starts transmitting near 2 μm, and the detector window usually cuts on near 7.5 μm, so most of the solar energy, in the visible and near infrared, never reaches the focal plane. Focusing the full solar spectrum at f/1 with τ = 0.9 would put about 10³ W/cm² in the spot; that is a visible-lens number. FLIR’s estimate for an uncooled camera is about 2 W/cm² on the detector for the Sun in the 7.5–15 μm band, the same order as the continuous-exposure limit they publish (do not exceed 2 W/cm²; 100 mW/cm² is the low-risk level). In-band irradiance still depends only on in-band radiance, F-number, and transmission, not on focal length. Focal length sets how many pixels the image covers and how much power sits in the spot. At 2 W/cm², the 465 μm image of a 50 mm lens has an area of about 0.0017 cm² and about 3 mW in band; at f/1.2 the irradiance falls as 1/(F#)² to about 1.4 W/cm². A longer lens burns more pixels, so the mask diameter cannot be copied from a short lens. A short look at the Sun often leaves a recoverable latent image on vanadium oxide and a permanent burn more readily on amorphous silicon; recovery time grows with exposure.

For the algorithm implementation, the geometry supplies four parameters:

  1. Masking radius: radius = 0.5·d/p × safety factor (1.5–2, to cover the diffraction halo, alignment aberrations, and scatter), with a feathered transition at the mask edge. From Table 6: 12 μm with 19 mm requires masking a diameter of about 15 px, 12 μm with 100 mm about 78 px, and 8 μm with 100 mm about 116 px.
  2. Position estimation: image-side detection of saturated bright clusters — the solar image saturates a core with a scatter halo around it — or attitude-side mapping of the Sun's direction onto pixel coordinates using an IMU and clock. The two routes back each other up.
  3. Mitigation: neighborhood interpolation or frame freezing over the mask, shortened integration, shutter insertion, a mechanical baffle for extreme cases, and a "do not point at the sun" user prompt.
  4. Reflected secondary images: sunlight reflected by water, glass façades, or car roofs keeps the same angular diameter with somewhat reduced irradiance and can appear anywhere in the frame — astronomy-based estimation alone will miss it, so image-side detection cannot be omitted.

The 384×288 format adds a proportion problem: at 100 mm the solar image spans 20% of the frame width (12% on 640×512), so masking costs more usable image area and the mitigation strategy needs to be finer-grained than on larger arrays.

Conclusion

The parameter relationships of an infrared lens compress into one chain: focal length × sensor format → field of view (observation area); focal length × pixel pitch → IFOV → observation range; F/# × pixel pitch → diffraction matching; F/# × focal length → entrance pupil and total solar power. Pixel pitch is the scale factor of the whole article — every switch between 17/12/8 μm shifts every table above by a full column: field of view and range scale in inverse proportion at fixed focal length, and the diffraction budget is re-allocated. Engineering selection can follow the chain: fix the range requirement and invert D = L·f/(n·p) for the focal-length class; check the field of view and stabilization plan for that class; look up the diffraction MTF to bound the F-number for the chosen pitch; and finally re-check the sun-burn protection parameters against the masking budget. Field of view and range always trade off within a single lens, and the balance point is set jointly by the sensor format and the mission definition.

Back to Tech Blog

Need selection advice or a custom solution?

Link-Card provides thermal module selection, secondary development and full-system solution support.

Our sales team will reply to the email or phone number you leave. You can also email sales@link-card.com directly.