Knowledge base · Optics & lighting

Optics and lighting that make the camera work

A Computer Vision knowledge guide for engineers and machine builders. How FOV, working distance and lighting decide contrast, before you order hardware.

Knowledge

Lens and lighting belong together

The camera sensor only sees what the lighting creates and the lens delivers. More megapixels will not recover contrast that never reached the sensor. A sharp lens will not fix a field of view that does not fit the machine.

Lighting creates the signal

For inspection, measurement and code reading, contrast between feature and background is the signal. Light geometry (backlight, bright-field, dark-field, dome) matters more than LED wattage.

The lens maps the scene onto the sensor

Focal length, working distance and sensor format set FOV, magnification and how much light reaches each pixel. Stop down for depth of field and you need more light to compensate.

Same FOV, same coverage

If the lens sees 200 mm but the light only covers 150 mm evenly, the edges fail even when the centre looks fine in the lab.

Selection

First decide what the requirements are

Write down what the application needs, not which camera model you already have in mind. Focal length follows from sensor format, FOV and working distance. Not the other way around.

Parameter	Why it matters	Typical trap
FOV	How much of the part must be in the image	A short lens “to be safe”, and too little resolution on the feature
Working distance	How much space there is between lens and part	Fix the mount first; later no lens fits
Smallest feature	Pixels across a defect, edge or code module	Megapixels on paper, without a lens that delivers them
Surface	Matte, glossy, curved, transparent or mixed	One ring light for every part family
Motion and trigger	Line speed, exposure time, strobe sync	Continuous light with long exposure: motion blur blamed on “the camera”

Rule of thumb in machine design: aim for roughly a 4:1 working-distance-to-FOV ratio when you can. It is not universal, but it prevents many redesigns because there is “no space left for optics”. More detail: resolution, FOV and working distance.

Practical example · Machine builders

Surface check on brushed aluminium in a packaging cell

A machine builder integrates an in-line camera after a filling station. The part is a brushed aluminium housing: mostly diffuse, with local glossy streaks. You need to see a shallow scratch and read a laser-etched batch code. Cycle time leaves little room for long exposures.

What the station needs

A FOV slightly larger than the housing face, with a fixed working distance from the machine frame.
Enough pixels across the etch and the scratch width. The lens must deliver that resolution, not only the camera.
Lighting that keeps glossy streaks from washing out the etch, while still showing the scratch as an edge or intensity change.
A strobe synced to the camera trigger, so short exposure freezes motion on the conveyor.

How Sedeco typically approaches it

We start from samples and the allowed working distance and FOV in the frame. For scratches on metal we often try dark-field or low-angle bar lighting. If the etch needs flatter illumination, we add a second geometry or a dome. Only when the image works on the real part do we lock focal length and product family.

Keep reading: inspection lighting · code reading & DPM.

Field reality

What often goes wrong in practice

Most failed Computer Vision cells do not have a “bad camera”. Optics or lighting was chosen too late, or chosen independently.

Camera and mount first, optics last

The bracket is welded, working distance is fixed at 80 mm, and only then someone asks for a lens that covers a 120 mm FOV on a large sensor. The short lens that hits the FOV may vignette or distort; the longer sharp lens does not fit. Redesigning the frame then costs more than selecting the lens with the mechanical space from the start.

Bright-field ring light on glossy metal

A white ring light is ordered with the camera because it is the default accessory. On glossy or machined metal it creates hotspots that saturate the sensor. Defects and DPM codes disappear into the reflection. Dark-field, dome or carefully angled bars belong in the first trial, not a software threshold on a burned image.

Stop down for depth of field, forget the light budget

Parts vary in height, so the iris is closed. Exposure climbs; on a moving line the image blurs. Continuous light is turned up and still fails under changing ambient light. What is missing is a strobe with enough optical power at the closed aperture. That is a lighting decision, not a “faster camera”.

Definition pages