Hi,
I have done some test to pick donuts but the problem is that the hole (size and position) and the form (ovality) of the donuts is changing al the time. So I have made 2 vision model one looking the hole for the x,y position and one model looking the form to search the correct pick orientation.
The problem is that if the product is not completly in the picture I reconise for example only the hole but not the form (see picture) . I my case it is really important to detect both picture at the same time. I have try to use the different window limits but I always got the problem.
Any idea would be grate? I do not know if it is possible to do this with the standard Pickmaster.
BR,
Fabrice
Hi,
To detect the product you could use Patmax or blob. Even though the product is pretty much circular you still want a picking angle, correct?
-If you use Blob and search for the black circle you should get a good approximation of the angle as long as there is some difference between its length and width.
-If you use PatMax I would suggest you train a couple of different models of varying shape and size. Probably you should just mask out the hole and only look at the exterior shape. The position filter in the position source can be adjusted so that you only get the target from the best model hit sent to the robot.
Now to the problem with products at the border of the image. To make sure you only process fully visible products in the image you can increase the trigger rate so that you are sure to capture the whole product in at least one image. Normally the auto trigger distance should calculate this for you, but you can adjust it manually. The second thing you need to do is to allow the product to enter fully into the picture before the coordinate is calculated. For this you can use “Limit position region” in the model definition dialog. The whole image is processed but only objects with a hit within the window are accepted.
Hi Frederik,
I am using of course the blob detection, this is the best way to do with such kind of product but the problem is that I could never get all items on the same picture within the different limit position region.
In the picture bellow I am looking for the hole and I found only one because the other one is on the edge of the limit search region.
In the picture bellow I am looking for the form of the donuts and due to the big area tolerance pickmaster is still recognize the product but reject it because on edge.
The biggest issue in this case is that the product can variate in size and form during production there for I need a big tolerance on the area and due to this it seems that the product is reconignized but rejected because on edge.
It is really difficult to ajust the right limit position region.
Any other suggestion?
BR,
Fabrice
Hi,
I guess you are/will be using a conveyor belt to transport the products, correct?
You are already using “No blob on edge” so that you don’t get an accepted hit until the product has entered fully within the picture. Then I guess the remaining problem is to make sure you capture all donuts completely within the picture at least once? To do this manually increase the trigger distance if it is too long now and adjust the overlap filter so that you don’t get more than one hit from each product (if it is captured in two or more consecutive images).
About the “limit search region”: This actually crops the image. So if you are using “no blob on edge” then any blob that touches this border will be discareded. In your case I would only use it to mask out the outer parts of the image where the lighting fades out.
If you for some reason need to find products on the border of the image I would suggest
using Patmax with a high elasticity setting to accurately find partially visible products. The blob centre point is affected if the product is on the border.
Regards Fredrik