Mall entrance people-counting
A fixed 4–6 mm lens trimmed to the doorway gives the counter a stable scene and consistent pixels on the threshold; we resist a varifocal here because every zoom change forces a re-calibration of the counting line.
AI Camera Optics · Guide
AI Fixed vs Varifocal Lens for Analytics — Engineer's Guide
A practical guide to the lens decision that quietly drives AI analytics quality on Abu Dhabi sites — fixed-focal-length versus motorised varifocal, the focal-length maths, and the scene calibration that follows. Picking the wrong lens cannot be fixed by a smarter model; picking the right one makes a modest model perform well. We make the decision at survey, with measurements rather than guesswork.
Get a Free Quote Call +971 55 651 7843
ADMCC ApprovedEngineer-Led SurveySince 2018Honest Scope
How do you choose fixed versus varifocal for AI analytics?
It comes down to whether the scene is fixed or evolving — and whether you need site staff to ever change the framing. A fixed-focal-length lens (commonly 2.8 mm, 4 mm, 6 mm, 8 mm or 12 mm on a 1/2.7" or 1/1.8" sensor) is the right choice when the scene geometry is locked: a doorway entrance counter, a fixed-perimeter wall, a corridor at a known distance, an LPR shot at a measured stop line. The optics are sharper at the chosen focal length, the lens module is more robust, and the AI analytic can be calibrated once and left to run. A motorised varifocal (typical 2.7–13.5 mm, 7–22 mm or 8–32 mm ranges with motorised zoom and focus) earns its place when the scene is still being defined, when the operator wants the option to re-frame after commissioning, when phased construction will shift the boundary, or when one camera must serve as both an awareness overview and an analytic tight shot at different times of day. The trade-offs are real: varifocals are heavier, drift in focus over months of thermal cycling, and demand re-calibration of the analytic when the zoom is changed. We compute the focal length at survey from the operational distance and the required pixels-on-target — for example, a 4 mm lens on a 1/2.7" sensor delivers roughly 80 horizontal-degrees FoV, which gives usable person-classification pixels out to about 12 m; pushing the analytic to 25 m on the same scene means an 8–12 mm focal length and a tighter FoV. We document the calculation on the layout and avoid the all-purpose 2.8 mm wide-shot that fails analytic work on every site. Multi-camera tracking only works when each camera's lens has been chosen against its zone. Call +971 55 651 7843.
The lens mistakes that quietly kill AI analytics
The most common error is the default 2.8 mm wide-angle camera repurposed for analytics — the FoV swallows the scene, pixels-on-target collapses at any useful distance, and the analytic spends its life rejecting low-confidence detections at the edges. A tighter focal length, even on the same camera body, would recover the analytic. The second pattern is a varifocal commissioned at the wide end then never re-tuned, defeating the whole point of buying the optical range.
The third is mismatched lens iris between day and night — a fast f/1.4 lens reads better in low light but throws depth-of-field away in bright sun, which can soften plate-readable LPR shots; we choose iris against the lighting profile, not the brochure. Every lens decision is logged. Call +971 55 651 7843 or request an optics survey.
- Focal length computed against operational distance
- Pixels-on-target validated before lens ordering
- Fixed-lens preferred for locked-geometry analytics
- Varifocal for evolving scenes or re-framing options
- Iris and depth-of-field matched to lighting profile
- Re-calibration scheduled after any varifocal change
Where the lens choice decides the outcome
Use cases for AI lens selection in Abu Dhabi
Perimeter wall and compound
A motorised varifocal (typically 2.8–12 mm) earns its place during commissioning so we can tune the framing against the wall edge, then we lock the zoom and document the focal length on the AMC record.
LPR at gate or stop line
An LPR shot is a precision job — fixed focal length chosen to put 130–150 px on the plate at the read distance, narrow FoV to reach that pixel count, a fast shutter to limit motion blur and a controlled IR profile; a varifocal that drifts in focus costs read rate.
Industrial yard awareness plus analytic
One camera cannot do both well — we prefer a wide awareness camera paired with a tighter analytic camera on the same pole, rather than a varifocal that compromises both jobs in the same shot.
At a glance
Fixed vs varifocal lens for AI analytics: side-by-side
| Factor | Fixed lens | Varifocal lens |
|---|---|---|
| Field of view | Set at manufacture | Adjustable focal length on install |
| Install flexibility | Lower — must choose the right lens up front | Higher — tune the field of view on site |
| Cost | Lower | Higher |
| AI analytics fit | Works when the scene matches the lens | Lets you size pixels-on-target for reliable analytics |
| Best for | Locked scenes — corridors, doorways, counting lines, LPR stop lines | Scenes still being defined — perimeter walls, phased sites, mixed awareness and analytic roles |
Request a free consultation and we will recommend the right fit for your Abu Dhabi site.
Common questions
FAQs
Is a varifocal lens always better than a fixed lens?
No — varifocals add flexibility but they drift in focus over months, are heavier and demand re-calibration of the analytic whenever the zoom changes. Fixed lenses are sharper at their design focal length and more reliable for locked-geometry analytics like doorway counting or perimeter walls.
What focal length suits AI perimeter analytics?
It depends on the wall distance and sensor size. A rough guide on a 1/2.7" sensor is 4 mm out to about 12 m, 6 mm out to 18 m and 8–12 mm out to 25 m for person-classification pixels-on-target; we compute the exact number at survey rather than installing by habit.
Why is the default 2.8 mm lens a bad choice for analytics?
It is a wide FoV of roughly 100 degrees or more in most modules — fine for awareness, but pixels-on-target fall away quickly with distance. By the rule of thumb above, a 2.8 mm lens reaches only about 8 m for person classification, vehicle classification holds out somewhat further, and LPR is unrealistic at any useful range.
Do I need to re-tune the AI analytic after a varifocal zoom change?
Yes — the ROI polygon, the IOU threshold, the perspective calibration and any counting line move with the framing. A casual zoom change without re-tuning is a common cause of analytic regression on inherited sites; we log every motorised change.
Do you survey the optics before quoting?
Yes — the free engineer survey precedes the itemised written quotation. We measure target distances, choose focal length against pixels-on-target, decide fixed vs varifocal on operational grounds and document the choice on the layout. Call +971 55 651 7843.
Ready to secure your property?
Get a free site survey and no-obligation quotation from Abu Dhabi's ADMCC-approved security specialists.
Need a free site survey or quote in Abu Dhabi?
Talk to Emirtech's engineers — we reply the same working day across Abu Dhabi and Al Ain. CCTV, security, IT, AV and smart-building solutions.
Mon–Sat 8AM–7PM · [email protected] · Since 2018 · ADMCC-approved for security systems