Before powder goes in and the lid goes on, one camera has to prove that a plastic scoop sits in the can. The can is the hard part: a specular funnel of thin metal that returns a point source as glare and shadow, while the matte scoop absorbs it. This guide gives the geometry that wins, the size it needs, the lens math for a 100 mm opening, and the fallbacks for lines with no room above the can. You leave with a parts list for the coaxial diffuse set-up and a pre-order checklist that works for scoops, desiccant sachets, lids and seals alike.
Why do ring and spot lights fail inside a metal can?
A ring or spot light fails because it is a point source: every ray reaches the can bottom from one azimuth, so the bright sector and the dark sector trade places with every scoop orientation, and the matte plastic sits at a different brightness from the metal whatever the exposure.
The can hands you three surfaces with three reflectances: a rolled bottom, bright drawn walls, and a matte plastic scoop. A ring light mounted round the lens, or a spot fired from one side, sends every ray into the can from a single azimuth. The wall that happens to lie at the bounce angle returns the source straight into the lens and saturates white; the wall opposite gets nothing and stays black. Move the can ten millimetres or rotate the scoop, and the hot spot travels with it.
You can see the scoop in such an image by eye; a threshold cannot follow a highlight that moves between frames. The failure is differential reflectivity stacked on non-uniform incidence, and exposure tuning cannot fix incidence. Bright field vs dark field lighting: the law of reflection decides shows the reflection law that puts the highlight where it lands, and lighting geometry for image contrast shows why camera, light and part must be picked as one triangle.
Which lighting geometry lights the can bottom evenly?
Full bright-field diffuse light through a coaxial port lights the can bottom evenly: a diffuse panel behind a half-mirror floods the field from every azimuth at once, so the bottom, the walls and the scoop all receive light of nearly the same intensity from the camera's own direction.
A coaxial diffuse head is a wide, flat panel firing into a half-mirror set at 45°; the camera shoots through the clear port in that mirror. Because the panel is wide and diffuse, every ray path is folded onto the lens axis, so the metal cannot find a bright bounce anywhere and the matte plastic scatters its normal share back. Tilt the can, shift it, spin the scoop — the brightness of each surface stays put, and that is what makes the check independent of how the scoop lies.
The half-step is a dark-field ring at 45°: its rays strike the walls and scatter down onto the bottom, far more uniform than a bright-field ring, but the light is still azimuthal, and the scoop's shadow swings round as the scoop turns. Good enough for a part that always lands the same way; not good enough for a flung scoop.
Full bright field comes in two forms: multi-directional from a dome lights hemisphere, or unidirectional from a wide coaxial panel. The dome form is covered in dome lights for shiny, curved and wrinkled parts, but a dome has to engulf the part, and a 110 mm deep can hides its own bottom from the dome's oblique rays. The flat port showers the bottom from the camera's own direction, which is what a can wants.
How large should the coaxial light be?
Size the emitting area larger than the field of view, not merely equal to the can opening: the reference build for this exact task used a 152 mm (6 in) square coaxial diffuse head, roughly 150 mm tall, over its can, and a smaller source leaves the field edges dark.
The mechanism behind the rule: the half-mirror only returns light that lands on it, so the emitting panel must overfill the mirror and the lens aperture, edge to edge. A panel the size of the opening vignettes the rim, and the rim millimetres are exactly where a flung scoop handle lands.
The coaxial lights range steps through the same ladder:
| Specification | LS-CXL-0192 | LS-CXL-0067 | LS-CXL-0027 |
|---|---|---|---|
| Cable length (mm) | 500 | 500 | – |
| Voltage | – | 24 V DC (12 V on request) | 24 V |
| Emitting surface (mm) | – | 102 x 100 | 60 x 60 |
| Price | $739.00 | $499.00 | $269.00 |
Values as printed in each product's data sheet; open a part for its drawing and full specifications.
The LS-CXL-0192 is a 140 × 140 mm coaxial light at 525 nm. For the 100 mm opening of the worked example, that panel sits 20 mm wider than the 120 mm field of view (140 − 120 = 20 mm) and 40 mm wider than the opening itself, so the rim stays lit even with the can off-centre. A 102 × 100 mm head takes openings up to about 80 mm; a 60 × 60 mm head suits closures and small cans near 40 mm.
How much headroom does a coaxial diffuse head need?
Budget a coaxial head about as tall as its own width — close to 150 mm of Z for the 152 mm square class — plus the camera above it; when that height does not exist, drop to a compact coaxial head or a thin diffuse panel with a viewing port.
Plan the stack from the rim up: 100 to 200 mm of air between rim and port — 190 mm in the worked example below (300 mm working distance − 110 mm can depth) — then the head's own height, then camera and lens above the port.
Where the gantry will not give you that, the compact coaxial is the first fallback. The LS-CXL-0036 is a high-precision coaxial light in a 74 × 50 × 50 mm envelope, one third of that height, with the same beam-splitter geometry over a smaller field. Thin diffuse panels with a coaxial viewing port are the second low-profile option; they reach a similar but softer result and suit flat foils and lids better than a deep can. More low-profile options are compared in compact machine vision lighting for tight spaces.
- Size
- 50-100 mm
- Voltage
- 24 V
When nothing at all can sit over the opening, swap geometry rather than shrink it: a focused head from the spot lights range, fired from an angle off the axis. The LS-SPT-0037 carries a 21° lens and a 99.9 × 141.6 mm lit area. A 21° full-cone beam at 300 mm covers 2 × 300 × tan(10.5°) ≈ 111 mm, enough for a 100 mm opening, and the head mounts on a bridge beside the line. You pay for that access with azimuthal contrast, so hold the part presentation steady or use a rotation-tolerant algorithm.
- Size
- 50-100 mm
- Voltage
- 24 V
What camera and lens cover the opening and the scoop?
With a 100 mm opening, ±10 mm placement, a 120 mm field of view, a 1600 × 1200 camera with 4.5 µm pixels and a 300 mm working distance, the lens is a 16 mm focal length: it covers 128 mm at 80 µm per pixel on the bottom.
Work the numbers in this order.
- Field of view: a 100 mm opening with ±10 mm of placement needs 100 + 2 × 10 = 120 mm.
- Sensor: a common 1600 × 1200 format with 4.5 µm pixels spans 1600 × 4.5 µm = 7.2 mm; substitute your camera's figures.
- Magnification and focal length at the 300 mm working distance:
That gives m = 7.2 ÷ 120 = 0.060 and f = 300 × 0.060 ÷ 1.060 = 16.98 mm, so fit a 16 mm lens. Confirm it: m = f ÷ (WD − f) = 16 ÷ 284 = 0.056, and the field becomes 7.2 ÷ 0.056 ≈ 128 mm, 8 mm wider than required, which absorbs a can sitting high in its nest. The pixel footprint on the bottom is 128 ÷ 1600 = 0.080 mm, so an 80 µm pixel maps a 15 mm scoop handle onto 15 ÷ 0.080 = 187 px — about 190 px of part width for the decision to bite on.
Depth of field decides the f-stop:
At f/8 with a two-pixel circle of confusion (2 × 4.5 µm = 9 µm), DOF ≈ 2 × 8 × 0.009 × 1.056 ÷ 0.056² ≈ 48 mm. The can is 110 mm deep, so focus on the scoop plane at the bottom and let the rim go soft; the rim never enters the decision. Closing to f/11 buys 48 × 11 ÷ 8 ≈ 66 mm if the wall must be readable too. Treat the formula as an estimate for choosing the stop, not a sharp boundary.
Run your own numbers in the lens focal length calculator.
What gray levels and exposure make the decision reliable?
Set the exposure so the bare can bottom lands near gray 200 on an 8-bit camera and the scoop reads at least 50 levels below it, then strobe the light for every frame; that pair of settings keeps the decision stable through can-lot and scoop-colour variation.
Gray 200 on a bare bottom leaves 55 levels of headroom under saturation (255 − 200 = 55), enough to absorb the wall's specular hits without clipping. Measure on a 20 px patch on the scoop and a second patch beside it on the metal; the gap between the two means is your margin, and 50 levels is the floor. Robust machine vision lighting: why adequate contrast is not enough explains what that margin protects across a season of can lots.
Hold the light, not the ambient: strobe the head from a controller on every frame and gate the camera exposure to the strobe, so the only light in the image is yours. Overdrive strobing without visible flashes keeps a strobed station comfortable to work beside.
With the margin held, the pass/fail logic becomes a validated-recipe problem, the same discipline described for presence detection with validated part recipes on the control side.
Which light do you choose for each line constraint?
Choose by the constraint above the can: full height and top access means the large coaxial diffuse head; top access but low Z means the compact coaxial; no top access at all means an off-axis focused spot; and a part you can stage in the open means the dome.
| Constraint above the can | Geometry that wins | Part to quote |
|---|---|---|
| Camera looks straight down, height is free | Coaxial diffuse, panel wider than the field of view | LS-CXL-0192 |
| Top access, but tight vertical space | Compact coaxial over a small opening | LS-CXL-0036 |
| Nothing can mount over the opening | Off-axis focused spot from a bridge | LS-SPT-0037 |
| Part can be staged out of the can | Multi-directional dome | LS-DOM-0085 |
The dome row covers the scoop itself on a stage before it drops in, or any curved shiny part you can present in the open. The LS-DOM-0085 is a Ø166 mm dome light with a Ø116 mm emitting area, lit from every direction a curved surface can turn towards. For the wider decision — surface finish first, feature second — read machine vision light geometry selection by surface and feature.
- Size
- 100-200 mm
- Voltage
- 24 V
What should you check before ordering?
Before ordering, check six things: panel wider than the field of view, head height inside your Z budget, lens covering the opening at your working distance, strobe wired to a controller channel, gray separation held across can lots, ambient light shared with the strobe.
- Opening diameter and placement tolerance — together they set the field of view, and the field of view sets the panel size.
- Height above the rim — a head of the 150 mm class wants that height again for itself, plus camera and lens.
- Can depth against depth of field — focus at the scoop plane and budget f/8 to f/11.
- One controller channel per station light, gated to the camera's exposure.
- Gray separation measured on the worst can lot you can get, against the 50-level floor.
- Ambient light at shift change — strobing defeats it only when camera and controller share the same pulse.
Then start at the light step of the system configurator with your opening and headroom numbers, browse the lights finder, or send a can and a scoop through the contact page and we will light them and send back sample images. More lighting decisions are collected under the lighting topic.
Questions
Can a dome light replace the coaxial head for this can inspection?
Only if the part can be staged under it. A dome is full bright field too, but multi-directional: its oblique rays hit the can walls, not the bottom, and the dome must physically engulf the part, which a 110 mm deep can prevents. Use the dome for the scoop presented on an open surface and the flat coaxial port for anything inside the can.
Why must the emitting panel be wider than the field of view?
The half-mirror only returns light that lands on it, so a panel the width of your field leaves the mirror's corners and the lens aperture partly dark. The dark corners of the field are exactly where the can rim and wall sit, and a dark rim can read as a missing scoop edge. Wider panels cost head height, which is the real trade.
What exposure freezes a can moving on a conveyor?
Keep motion blur under one pixel: at 80 µm per pixel and 0.5 m/s belt speed, exposure must stay below 160 µs, because 500 mm/s × 160 µs = 0.08 mm. Strobe the coaxial head from a controller for 100 µs per frame and blur drops to 0.05 mm, comfortably inside the pixel.
Does the LED colour inside the can matter?
Yes, with a monochrome camera, because the plastic scoop and the metal bottom have different reflectance curves: a single colour widens or narrows the brightness gap the camera records. White is the safe first choice and the only one with a colour camera. If white separation is marginal, trial a green head and a blue head before you change geometry; the gap often moves more with wavelength than with exposure.
How do you prove the set-up before release?
Rotate and flip the scoop in a can and log the gray separation for every pose; then repeat with three can lots and two scoop colours. Every pose must keep the 50-level margin with the worst lot. A pose that fails points at the geometry — asymmetric light, not a wrong threshold — and the fix is panel width or a longer exposure, in that order.


