How do multiple checks work in video egg inspection?

What is two-of-three voting in egg inspection?

Two-of-three voting is the rule that decides, on video, whether an egg is rejected. Vigiovo measures each egg three times while it passes the camera. If the defect appears in at least two of those measurements, the egg is rejected. If it appears in only one, the egg is not rejected for that reason.

Each measurement uses the same fixed limits as the rest of the system. Voting does not change the dirt limit or the crack limit. It decides how many times the criterion must be met before a rejection counts.

Why measure the same egg more than once?

A single measurement can be wrong because of the moment it was taken. Strong glare on the shell, the shadow of a neighbouring egg or a change in position can make an egg look dirty or cracked in one frame and normal in the next. A one-frame rule would turn that passing error into a rejection.

Measuring more often costs processing, so Vigiovo separates the cheap work from the expensive work. Colour segmentation, which separates shell from background, runs on every frame. Defect measurement, the heavy part, runs only in the three checks of each egg.

How are the three checks spaced along the conveyor?

The checks are spaced by how far the egg travels, not by time. Each measurement happens at least half a radius of travel after the previous one, so the three measurements correspond to different positions of the egg. The scheme resembles the 1/3, 2/3 and 3/3 checks on a quality panel.

To know that the three measurements belong to the same egg, the system follows each egg from one frame to the next. It uses nearest-neighbour matching, a constant-velocity prediction and a deterministic rule for ties. This keeps each egg's identity from the start to the end of its passage.

What does voting not solve?

Voting removes noise from a single frame, but it does not create information. If the camera cannot see a defect, no number of checks will reveal it. Hairline cracks, the air cell and internal blood spots are beyond the reach of a surface camera, and voting does not change that.

A defect that appears in only one of the three checks is not rejected either. So if the line has strong glare, the fix is to review the conveyor lighting, not to rely on voting to hide the problem.

How do you verify voting on your line?

Before you rely on video classification, test the voting with eggs your team already knows. The test depends on your line, not on a fixed number. It complements calibration with labelled eggs.

1. Set aside a batch with clean, dirty and visibly cracked eggs, already graded by your team.
2. Film the batch passing along the conveyor under production lighting.
3. Compare the video rejections with your team's grading, egg by egg.
4. Note the eggs that changed class between one frame and the next and investigate the cause, such as glare or shadow.

Frequently asked questions

Does two-of-three voting delay the decision on each egg?

Yes, slightly. A decision on an egg is only ready after the third check, so it comes after the egg has travelled the distance between the measurements. For the processing time of the system, see the article on latency and throughput.

Can two-of-three voting let a dirty or cracked egg through?

It can let through a defect that appears in only one of the three checks, for example if the light fails for an instant. That is the trade-off of the rule: fewer rejections caused by glare, with some risk of missing a very fleeting defect. Reviewing the lighting reduces both problems.

Why does Vigiovo use three checks and not more?

The number of checks is fixed in the method: three measurements per egg, with rejection on two. Each defect measurement costs processing, and the heavy part is done only those three times to keep inspection in real time.

Sources

1. Como o Vigiovo classifica ovos sem IA / How Vigiovo grades eggs without AI, Vigiovo