Capgemini, a world leader in consulting and technology services, interviewed 300 companies regarding AI and computer vision. In their report “Scaling AI in Manufacturing Operations: A Practitioners’ Perspective”, one aspect, in particular, caught my attention, here’s what it is:

Why small and medium companies don’t invest in computer vision and AI like big corporations?

Certainly, COMPUTER VISION and AI help manufacturing companies but many small businesses seem to be scared of this type of change, in my experience, there are 3 main causes:

  1. The main cause is the costs. Many business owners think that the expense is very high.
  2. The time, the fear that one’s business may be stopped for days or weeks
  3. There are not many professionals able to understand the needs of the company and able to offer customized solutions.

To answer these fears I will tell you a success story in the field of computervision of a few years ago.

Computer vision and AI applied to cucumber production

This story comes from Japan, a boy who worked as a system designer and their parents produced cucumbers. The problem with this production was the classification of the cucumbers, the product with fewer defects had a higher value than a cut or deformed cucumber.

Only Makoto’s mother had the knowledge to classify cucumbers into 9 different qualities and teaching it to other people would take a long time. To solve this problem he decided to use computer vision and artificial intelligence.

Makoto has created a dataset of 7000 photographs of including and classifying all categories of cucumbers. At a first check she had obtained a match result of over 90% but in reality the accuracy dropped to around 70%.

As you can see from the photo, he built a machine using a raspberry pi for the control. Let’s remember that all this happened a few years ago when artificial intelligence with the use of frameworks was still in its beginning. Despite this, productivity has increased, guaranteeing valid help.

Example of Computer Vision and AI implementation in a small company

Let’s start with the question: “How COMPUTER VISION and AI can help manufacturing companies?” To answer, I try to create a business solution from scratch for a company that does not exist: Caps l.t.d


The company produces golden-colored caps but the production is not always perfect. For these reasons, an employee checks the caps one by one to discard the damaged ones.

Human control is not ideal because it is slow and it is easy to make mistakes so you have to find an automated solution to solve the problem. I can divide the solution into 4 steps

1. Caps detection

The first stage is the identification of the object. The software must identify the object on the conveyor belt.

As can be seen from the photo, all identified objects have a green outline.

2. Caps tracking

Once the object has been identified, it must also be tracked and assigned an ID that allows it to be uniquely identified.

This allows you to keep the position of the object concerned and possibly make a robotic arm, gate or other intervene for removal.

3. Caps classification

This is perhaps the most complex phase of the project because the software has to understand when the product is good and when there are problems. For this step I have written an artificial intelligence model for this exact purpose.

3.1 Caps training

It is not enough to write the model but it is essential to teach the artificial intelligence what it has to find. To do this, thousands of photos of each category have to be taken, considering, among other things, also the exposure of light. The more photos there are, the higher the success rate.

In the photos there is a small example of what should be done. All caps have been divided into 3 categories: casps ok, color damage and shape damage.

3.2 Run the model

After training and making sure that the result is close to 100%, all that remains is to integrate the model with the rest of the code and get it started.

As you can see from the photo, the software found the problem in a cap and reported it correctly. The object being traced, it is also possible to obtain the coordinates and act accordingly.

4 Associate an action

This is not the purpose of this demonstration but the logical consequence is to associate an action with the work of the software. For example, stop the conveyor belt, trigger an alarm or send the coordinates to a robotic arm.