Camera selection for egg defect inspection on rollers

An egg on a roller sorter is a friendly target: a pale shell on a dark background, a known size range and a steady line speed. The hard part is picking the camera architecture before you spend money on the wrong one. This guide sizes the camera from the smallest defect you must catch, then decides area scan versus line scan, field of view per camera, line rate, shutter and spectrum for a multi-lane roller bed. The worked numbers assume six lanes at 65 mm pitch, eggs about 50 mm wide, and a line speed of 0.5 m/s; scale them to your machine.

How many pixels do you need across an egg?

Plan for about 0.33 mm per pixel: the smallest defect you will flag, a 1 mm dirt spot or crack, must span at least 3 pixels, which puts roughly 150 pixels across a 50 mm egg.

That 3-pixel rule is the whole sizing method. Eggs run about 45 to 55 mm wide and up to 60 mm long, so the defect budget — not the egg — sets the sensor. The same pixels-per-object logic drives camera resolution for defect detection and, on grains instead of eggs, line scan grain sorting.

Working the numbers: 1 mm ÷ 3 = 0.33 mm per pixel, and 50 mm ÷ 0.33 mm ≈ 150 pixels across the egg. Tighten the target to 0.5 mm hairline cracks and the budget halves to 0.17 mm per pixel, about 300 pixels across — still comfortably inside one sensor. Set the field of view first, then divide it by this budget to get the pixel count the camera owes you.

Area scan or line scan for a roller sorter?

Choose area scan when your sorter handles eggs in singulated stations and one frame per station is enough; choose line scan when eggs stream continuously down the roller bed and every lane must be imaged all the time.

Both architectures work on rollers, because the rollers spin each egg and sweep its shell past the view. An area camera at a station freezes a frame with a strobe and judges each egg once. A line camera builds the image line by line as the bed advances, and the egg's rotation turns its trace into a spiral that walks across the shell. One line camera serves every lane; area scan multiplies cameras with lanes but keeps triggering simple. Compare the two families on the area scan cameras and line scan cameras pages.

Station cameras want color for shell tint and blood spots, and a global shutter for the spinning egg; this 2.5 MP GigE model fits that brief:

A continuous bed wants color at modest bandwidth; this 4K color line camera reads 7 μm pixels across a 29 mm image circle on GigE:

What field of view should one camera cover?

One 2 to 3 MP area camera covers two or three eggs, a field of roughly 130 by 80 mm, while one 4K line camera covers four lanes of 65 mm pitch and an 8K camera covers six or more.

Size the field from the mechanics: lane pitch around 65 mm on roller beds, egg length up to 60 mm, so allow 80 mm along the travel and lanes × 65 mm plus margin across. Four lanes make a 300 mm field; a 4K line across it resolves 300 mm ÷ 4096 px = 0.073 mm per pixel, about 14 pixels on a 1 mm defect. Six lanes need roughly 420 mm, and an 8K line gives 420 mm ÷ 8192 px = 0.051 mm per pixel, about 20 pixels per millimetre. The lens must then cover the sensor width at that field — match image circle to line length with the line scan lens selection guide.

Two step-ups cover wider stations and full beds:

The MV-CH120-11UC raises an area station to 12 MP on a 1.1-inch sensor with a global shutter over USB3. The LS-LCM-0099 covers the whole bed in mono, 8192 pixels per line, over 10GigE.

What line rate keeps the image sharp at line speed?

Advance one line for every pixel footprint the web travels: at 0.5 m/s a 4K camera over a 400 mm web needs about 5.1 kHz and an 8K camera needs about 10.2 kHz.

Set the rate using how to set the line rate of a line scan camera and clock it to the encoder as the next section shows. For area scan the equivalent number is exposure: blur stays under one pixel footprint — about 0.08 mm in the two-to-three-egg station — only if the shutter closes within 0.08 mm ÷ 500 mm/s = 160 μs, a strobe number, not a gain number. The fps and shutter arithmetic for moving webs runs the same budget.

Depth needs one stop of attention. Eggs vary a few millimetres in height, and at the 0.07 magnification of a 400 mm field on a 4K, 7 μm sensor, f/5.6 holds roughly 33 mm of depth at a 2-pixel circle of confusion (depth ≈ 2·N·c·(1+m)/m²) — treat that as an estimate and confirm it on the first tray.

Do moving eggs force a global shutter?

Yes for area scan: a rolling shutter reads rows in sequence and shears moving objects, so spinning or travelling eggs come out skewed unless the sensor is global shutter or a strobe freezes the frame; line scan sidesteps the problem because each line is its own instant.

Rotation makes this worse than a flat belt: an egg spinning in the station moves shell material sideways through the frame while rows are still being read. Pick the sensor with the global versus rolling shutter trade-off in mind, and if a rolling-shutter body is already on the bench, shorten the exposure and add strobe power before replacing it.

Do you need color, or SWIR to see inside the shell?

Start with mono or color in the visible band plus transmitted candling light under the rollers: color earns its keep for blood spots and shell tint, and SWIR only earns its cost when subsurface cracks survive a proper candling trial.

Dirt and open cracks are surface features; mono plus the right lighting finds them, since a color sensor's Bayer mosaic buys no extra geometry. Candling is the classic second view: a bright light below the rollers makes each egg glow, blood and meat spots read as dark blobs, and cracks light up as thin bright or dark lines.

Some graders add a UV station after washing to check shell coverage — reflected UV versus UV fluorescence imaging separates those two modes. SWIR around 900 to 1700 nm penetrates the shell better for some subsurface faults, at higher sensor cost and lower line rates. Threshold rules catch gross defects; when stain variety pushes you to a learned classifier, bring it online in stages, as the AI vision inspection commissioning guide lays out.

How do you sync the camera to the rollers?

Trigger each area frame from a photoelectric sensor at the station, and clock every line scan line from a rotary encoder on the roller drive so pixel scale holds when line speed changes at startup, gaps or lane changeover.

Encoder math is one division: counts per millimetre = encoder counts per revolution ÷ roller circumference, and one line fires each time the accumulated counts equal the pixel footprint. A 2048-count encoder on a 100 mm roller gives 20.5 counts per mm; at 0.0977 mm pixels that is one line every 2 counts. The encoder ticks per pixel guide works the full example. Drive the strobe and candling LEDs from a lighting controller for line scan cameras so pulse width tracks the line rate instead of flickering with it.

What should you check before ordering?

Nothing here needs a trial installation to predict: write the defect size, count the lanes and read the top line speed off the mechanical drawing, and the camera choice falls out of three numbers.

• Defect spec in millimetres. 1 mm dirt means 0.33 mm per pixel; 0.5 mm cracks mean 0.17 mm per pixel. Every later choice follows from these two numbers.
• Lane count and pitch. Width = lanes × pitch + margin: 300 mm fits one 4K camera, 420 mm takes an 8K head or two 4K heads.
• Top line speed. Size the line rate at maximum speed, not average: the example's 5.1 kHz doubles to 10.2 kHz at 1 m/s, and the data rate doubles with it.
• Shutter and strobe. Area stations need a global shutter and a strobe under 160 μs at the example speed; line scan needs the camera's rated line rate to hold at your pixel count.
• Depth of field. Roughly 33 mm at f/5.6 in the example; stop down further if your eggs stand taller or the rollers crown.
• Sealing. A wet egg room is washdown territory: mount cameras and lights in sealed housings and buy them by both rating systems — how IP codes map to NEMA 4 and 4X.
• Sample tray. Run 100 eggs covering every known defect type through the whole chain before ordering per-lane hardware.

Which camera should you actually order?

For two to three lanes in station handling, order a 2.5 MP global-shutter color area camera per station with a strobed candling light; for a continuous bed, order the 4K color GigE line camera up to four lanes and move to the 8K 10GigE mono camera at six lanes or more.

Recut by case. A station sorter chasing 1 mm dirt and blood spots is done with one area camera per station plus candling below. A continuous bed up to four lanes runs the 4K color head at about 5.1 kHz in the example and still resolves 0.5 mm cracks, since 0.073 mm per pixel beats the 0.17 mm budget. Wider beds take the 8K head at about 10.2 kHz, with a separate 4K color camera on lanes where tint matters. Subsurface cracks get candling first and a SWIR trial second, on real eggs.

Size the exact sensor and interface in the camera finder, start a full build in the system configurator, or send a multi-lane drawing through the contact page and we will lay the lanes out with you. More worked inspection layouts live on the applications topic page.