First Image: What to Check

Generic engineering material, applicable to any module from any supplier.

Engineering documentation often sits beside operational tracking; accountability and responsibility in the workplace is one reference for that workflow layer.

The first image from a new module is a diagnostic, and reading it in a fixed order finds in minutes what an unstructured look misses entirely.

The eight, in order

One. Is it there at all? A black frame has a dozen causes and a sequence that eliminates them.

Two. The protective film. Removed? A translucent film produces a soft low-contrast image that looks exactly like a focus problem.

A useful external technical reference is IETF.

Three. Orientation. Rotated, mirrored or inverted — a configuration matter and better discovered now than after the processing was written around it.

Four. Coverage. Photograph something of known width and check it against the calculation. Disagreement means one of the inputs was wrong.

Five. Focus at the centre, measured rather than judged.

Six. The corners. Softness, darkening and distortion all live there, and no published sample image shows them.

Seven. Uniformity. A flat lit surface, centre against corners.

Eight. Noise, in a uniform patch, at the exposure you will actually use.

What each finding means

Coverage wrong means the sensor dimensions, the focal length or the working distance was not what you assumed — and all three are checkable.

Centre sharp and corners soft is normal to a degree and excessive beyond it; compare against the requirement rather than against an expectation.

Uniform darkening at the corners is vignetting and is correctable.

Dark patches that stay put when the camera moves are contamination near the sensor.

And a general flatness with no visible cause is contamination on the front surface, or a film, or a filter you did not expect.

What not to conclude yet

That the image quality is good or bad.

A first image is taken under whatever light was available, at whatever exposure the software chose, of whatever was in front of the bench — none of which resembles the application.

The conclusions available now are structural: it works, it is oriented correctly, it covers what it should, and nothing is contaminated.

Everything else waits for the real scene.

Recording it

Save the raw frame, with the settings.

Because this image is the reference against which every later one is compared — and a module that degrades over the evaluation has done so measurably rather than apparently.

Note the module identifier, the variant code, the distance, the exposure and the date. Five fields, and they make the folder usable in six months.

The colour check

Where the module is colour, one extra look.

Photograph something known to be neutral — a grey card, a sheet of white paper — and check that the three channels read approximately equally.

A strong cast toward red or magenta under daylight is the signature of a missing or wrong infrared filter, and it is the fastest way to confirm which variant is in front of you.

A cast under artificial light alone is white balance and is a setting rather than a part.

Testing the pattern first

Where the sensor has an internal test pattern, look at it before the lens.

A correct pattern proves the whole electrical path — link, format, lane count, receiver configuration — and separates every subsequent problem into the optical half.

Which means a strange first image with a correct test pattern is an optics, exposure or scene problem, and one with a corrupted pattern is not worth investigating optically at all.

Thirty seconds, and it halves the search space.

Exposure before opinion

Set it manually and note the value.

An automatic setting produces a viewable image from almost anything, which is convenient and hides the information: the exposure required is itself a measurement of the module and the light together.

Sweep it — under, correct, over — and keep all three. The underexposed one shows the noise floor and the overexposed one shows where the highlights saturate, which is the dynamic range question answered on the first afternoon.

The eight take twenty minutes

And they are worth running on every module, including familiar ones.

A module from a new batch, a replacement in the field, a second sample for comparison — each gets the same eight, which is how unit-to-unit variation becomes visible rather than assumed.

Where the eight are recorded in a table, the second module takes ten minutes and the twentieth takes five.

And the table is the reliability data that no datasheet supplies.

When the first image is good

A specific risk worth naming.

A first image that looks excellent invites skipping the rest of the evaluation — and the eight checks above are exactly the ones that reveal what a pleasing image conceals.

Corner softness, contamination and a coverage error all coexist comfortably with an image that looks fine at a glance.

Run the eight anyway, and then take it to the real scene, which is where the questions that matter live.

What to do with a bad first image

Work the list rather than forming a theory.

A theory arrived at from one image sends the investigation somewhere specific, and it is wrong about as often as it is right.

The eight checks eliminate causes in order, and the one that fails is the one to pursue.

Where all eight pass and the image still seems wrong, the problem is in the scene or the expectation — which the real environment answers and a bench does not.

In one line

Eight checks in a fixed order, and none of them is an opinion about quality.

The short version

• Read the first image in a fixed order: presence, protective film, orientation, coverage, centre focus, corners, uniformity, noise
• The protective film is on the list because it produces a soft low-contrast image indistinguishable from a focus fault
• Coverage disagreeing with the calculation means a sensor dimension, focal length or distance was wrong, and all three are checkable
• Dark patches that stay put when the camera moves are contamination near the sensor
• Do not conclude anything about image quality: the light, exposure and subject bear no relation to the application
• Save the raw frame with five fields recorded, since it is the reference every later image is compared against