Face recognition starts with face size in the frame
A terminal working from 0.3 to 1.5 meters sees a large change in apparent face size. The camera must keep enough facial detail at the far distance while avoiding excessive cropping at the near distance.
Instead of selecting a lens only by diagonal FOV, simulate or measure the horizontal scene width at several distances. Then check how many pixels a typical face occupies.
One fixed focus setting can cover more than expected-if the optics are chosen correctly
Face terminals often use fixed focus because the subject range is predictable and fast capture is valuable. A suitable depth of field can cover a broad distance range without waiting for autofocus.
Autofocus is useful when the distance range is larger or the camera is also used for close document capture, but focus speed and hunting should be included in the user-experience test.
Mounting height changes the face angle
A camera above a large display may look down at shorter users, while a lower camera may look upward at tall users. This changes eye visibility, perspective and the usable recognition zone. Mechanical placement should be optimized before compensating with an extremely wide lens.
Backlight and screen light are common failure sources
Entrance terminals may face bright doors or windows. Kiosks may illuminate the face from a large display. Exposure and WDR/HDR behavior should be evaluated with those real lighting conditions.
At night or in controlled access, IR illumination and an IR-sensitive channel can support certain liveness or recognition methods, but the exact requirement comes from the algorithm design.
Near and far tests should be treated differently
At 0.3 m, check that the face is not cropped and that lens distortion does not exaggerate facial geometry. At 1.5 m, check that eyes and key features still contain enough usable detail. Intermediate positions should confirm that focus and exposure transition smoothly.
Create near, nominal and far capture zones
A useful engineering test marks three floor positions, for example 0.3 m, 0.8 m and 1.5 m. At each position, record face height in pixels, headroom, focus sharpness and exposure. This turns a vague "supports 0.3–1.5 m" claim into visible evidence.
If the far face is too small, a narrower lens or different mounting position may help. If the near face is cropped, the system needs more coverage or a different user position.
Recognition tuning should not hide optical weakness
Algorithm thresholds can sometimes be relaxed to make a weak image pass, but that may reduce security margin or increase false accepts. The optical design should first produce consistently usable face images; algorithm tuning should then optimize around that stable input.
Define the acceptance metric with the algorithm team
The final camera choice should be based on recognition and liveness performance under the expected population and lighting, not on an arbitrary megapixel target. Image capture, preprocessing and algorithm thresholds should be evaluated together.
FAQ
Is autofocus required from 0.3 to 1.5 m?
Not always. A fixed-focus design with suitable depth of field may cover this range, depending on lens and sensor.
Should I use the widest possible FOV?
No. Excessively wide FOV can make distant faces too small and increase distortion. Use the angle needed for the capture zone.
Can one RGB camera handle liveness detection?
Some algorithms can, but others require IR or binocular channels. The camera configuration must follow the liveness method.
