Skip to main content
CerebraTech AI
Technical knowledge

Why camera inspection systems fail, and why it is almost always the lighting

Written by CerebraTech AI engineering teamAbout 5 min read

Teams starting out with camera-based part inspection almost always focus on picking the AI model first. But across many real installations, the problems that make a system unstable — missing defects at certain times of day, or getting less accurate after a few weeks — almost never come from the model. They come from lighting that will not hold still.

Why lighting matters more than you think

Every vision model learns from pixel patterns. If the light changes from what it was when the training data was captured — afternoon sun through a window, or the site lamps starting to degrade — the pixel values of the same part change even though nothing about the part changed. The model then makes the wrong call, and it looks like an AI accuracy problem when the real problem is upstream.

A rule of thumb that holds on nearly every project: if the misses cluster into a time of day (say 1pm to 3pm) or into one season, suspect the lighting before you suspect the model.

Light types used most often in industrial inspection

Light typeWhen to use itWatch out for
Bar lightFlat surfaces needing even light along a long axisControlling the reflection angle on glossy surfaces
Dome lightCurved surfaces, or uneven reflectivityCosts more than a bar light
BacklightWhen you need a silhouette to measure shape and sizeCannot inspect colour or surface finish at all
CoaxialHighly reflective surfaces such as polished metalDistance from the part matters a lot; it has to be exact

Do not mix ambient light (windows, factory lamps) with the light you install for the inspection system. Where you cannot avoid it, enclose the station to keep outside light out — do not simply turn the system light up to "drown" the ambient light, because brightness that drowns it at one angle will not drown it at another.

Checklist before you blame the model

Before touching the model or the confidence threshold, always check these three:

  1. Photograph the same part at different times of day (morning, midday, afternoon, evening) and compare average pixel brightness — a difference above 10–15% means the lighting is not stable enough.
  2. Check that the light enclosure, if there is one, really is sealed, with no gap for sunlight to get in.
  3. Confirm the lights are industrial-grade with high consistency, not ordinary LED lamps whose output degrades unevenly over time.

If all three check out and it still misses, then look at the model or the threshold.

In short

Lighting is the foundation of every vision inspection system. Investing in the right light set and enclosure up front is usually far cheaper than repeatedly re-tuning the model afterwards — for more on camera placement and working distance, see the vision inspection solution.

Want to talk this problem through with our team?

A free 30-minute conversation, no commitment, no jargon you cannot follow.

Talk to us for 30 minutes first

This site uses very few cookies

We use only the cookies necessary to remember your language choice and save this consent preference (your chosen theme is remembered via browser local storage, not a cookie). Our analytics tool uses no cookies and collects no personal data.

Read the cookie policy