When the System Doesn’t “See” Well: The Most Common Causes of Missed Defect Detection

The adoption of Computer Vision systems for industrial quality control is transforming the way companies inspect products, components, and processes. From automotive to food & beverage, from cosmetics to electronics, smart cameras now detect defects in real time with precision far beyond human capabilities.

And yet, despite the sophistication of modern algorithms, there are times when the system simply… doesn’t see. A missed detection can be costly: product returns, line stops, reputation damage, or even large-scale recalls.

Why does it happen? What are the most common causes?
Let’s break them down.

1. Uncontrolled lighting: the number one enemy of computer vision

Light is the “fuel” of any Computer Vision system.
Even a minimal variation can compromise the entire detection process.

Common issues:

• Glare and reflections on shiny or metallic surfaces

• Shadowed areas that hide defects or alter contrast

• Ambient light fluctuations, especially on unshielded lines

Why does this matter?

Algorithms rely on pixel differences. When lighting changes, pixels change too.
The result: the defect goes undetected… because it effectively “disappears.”

2. Inadequate or poorly calibrated optics

Even a high-quality camera can fail if the optical setup is not correctly chosen or maintained.

Typical mistakes:

• Insufficient depth of field, leaving parts of the image out of focus

• Perspective distortion on three-dimensional objects

• Resolution too low for the size of the defects being detected

Final effect:

If the image isn’t sharp, what the algorithm cannot see clearly… does not exist.

3. Incorrect camera positioning

The “point of view” is everything.
Even a perfectly designed system loses effectiveness if the camera is mounted in the wrong place.

Frequent problems:

• Wrong angle, hiding scratches, cracks, or impurities

• Line vibrations causing blurred images

• Distance variations caused by mechanical imprecision

Consequence:

The defect is there, but it is either outside the field of view or too distorted to be recognized.

4. Poor object positioning or lack of synchronization

When the object passes in front of the camera, it must be in the correct position at the correct time.

Typical causes:

• Lack of synchronization between sensors, PLC, and camera

• Line speed too high relative to the frame rate

• Mechanical misalignment of items on the conveyor

The risk:

The camera captures the wrong moment, and the defect “slips through.”

5. Algorithms not optimized or poorly trained

Modern Computer Vision often relies on AI and Deep Learning.
But if the model is not properly trained… it will see poorly.

Common issues:

• Incomplete dataset (too few examples of real defects)

• Unbalanced dataset (too many good samples vs. defective ones)

• Overfitting: the model works in the lab but fails on the real line

• Underfitting: the model is too simple to capture process variability

Effect:

The system doesn’t recognize defects it has never “seen” or cannot generalize.

6. Poor maintenance

Sometimes the problem isn’t technological but… mechanical.

What can degrade:

• Dirty lenses (dust, oil, micro-drops, vapors)

• Components shifting over time

• LED illuminators losing intensity

Result:

Image quality slowly worsens and detection performance drops until failure.

7. Material or defect variability

Not all defects look the same, and not all objects are identical.
Production processes with high variability demand more robust systems.

Examples:

• Surfaces that change color during processing

• New types of defects not included during training

• Materials that reflect light unpredictably

Why it happens:

The system was designed for an ideal scenario… which rarely exists in reality.

How to prevent missed defect detection

The solution is an integrated approach combining:

✔ Careful hardware design

(camera, optics, lighting)

✔ Algorithm modeling and validation

(intensive testing with real and simulated data)

✔ Continuous monitoring in production

with performance metrics and automatic alerts

✔ Scheduled maintenance

to preserve image quality over time

Conclusion

When a Computer Vision system fails to “see,” it’s almost never the algorithm’s fault alone.
Computer Vision is a complex ecosystem where lighting, optics, mechanics, software, and data must work in perfect harmony.

Understanding the most common causes of missed detection helps companies:

• increase inspection reliability

• reduce scrap and rework

• avoid line downtime

• improve perceived product quality

In an increasingly automated world, seeing well means producing better.

Do you want to know more about Computer Vision for Manufacturing Production Control? Write at info@metalya.it