machine vision: learning objects

Hello,
I just registered myself on this forum, I'm currently doing my graduation internship. The assignment involves machine vision.

I have searched online to find a machine vision system that can be teached without developer software. The situation is that the customer has to be able to teach new products without having access to the vision software.

I'm looking for a vision system that is able to add a class after a trigger from a HMI. I found several vision systems that are able to classify on predefined pictures (omron & cognex for example) but they have to be predefined in the developer software. If a customer has a new product (without the ability of QR/ID codes) then the product is not recognised and can only be added in the software.

Is this possible?

welcome to the forum.

I am not aware of any such product.
pattern recognition in any vision system requires that you take a typical part, clean it up, use that as the pattern, and give the software a +/- tolerance.

Cognex is my preferred choice.
I have also used Keyence.
james

Welcome to the forum,
Take a look at Omron machine vision.
Menu driven with no external software.


Look at the communication manual.

If the region of interest exists, you should be able to teach from the hmi.


Hope this helps.
gtsuport

gtsuport said:

Welcome to the forum,
Take a look at Omron machine vision.
Menu driven with no external software.


Look at the communication manual.

If the region of interest exists, you should be able to teach from the hmi.


Hope this helps.
gtsuport

Click to expand...

I have taken a look at Omron. I have found the FH series and the FHV7 smart camera. Is it possible with both?

In the sysmac software I tried the classification option. I want the operator to only add/delete classes and not be able to change other detections/inspections. Is that possible?

James Mcquade said:

welcome to the forum.

I am not aware of any such product.
pattern recognition in any vision system requires that you take a typical part, clean it up, use that as the pattern, and give the software a +/- tolerance.

Cognex is my preferred choice.
I have also used Keyence.
james

Click to expand...

Thanks,
Cognex does have the visionPro software that has the ability to call functions from a custom GUI. I found a video (https://support.cognex.com/en-nl/downloads/detail/visionpro/2689/1033) from cognex that shows a training button for the patmax function. Do you have experience with this? Because it looks like the GUI can be custom made so only the necessary functions are accesible for the operator.

Not sure what you mean by classes.


Any of the communication protocols will allow you to re-register the model data with the current image.

gtsuport said:

Not sure what you mean by classes.


Any of the communication protocols will allow you to re-register the model data with the current image.

Click to expand...

By classes I mean that multiple (different) object can be distinguished. The classification function in Omron does that. The re-registering sounds good if it doesn't overwrite the data but add an extra class in the classification function.

No it would not add any objects for detection.


It would only re-register the current image.


You could check with your Omron distributor.
They may have a system you could test with.


gtsuport