loading
 
Anonymous 8 years
Patrick,

But given the images above do all those comprise one group? I.e. what are you comparing these images to? Each other? Or other not shown images?

Or another way ... how many shape matching profiles are you matching against?

STeven.
patrick  [14 posts] 8 years
Steven,
For background info:
The part is a rubber product which tend to get mixed up during packing so the idea is we have a carousel doing various other jobs plus 3 cam vision checks to ensure the part on the carousel is the part that is to be packed.
I have a PLC which knows the part to be packed via a barcode reader.
I then have the rs232 on the  RR PC used to tell the PLC the best matches.
The PLC will be programmed to know which camera result is most reliable per product being packed.
Wrong parts are ejected off the carousel.
Two cameras (1&2)are doing shape matching on the side profile of the product with two libraries of 70 odd shapes/images.
Below are two images on camera 1 which I match against library1 and the next two are from camera 2 which I match against library 2
Ideally i would like to match the plan view image from this third camera (camera 3) with an image library. Pat



Anonymous 8 years
Patrick,

Any luck on getting those new images with altered lighting? If not let us know and we will see what we can do with the images above.

STeven.
patrick  [14 posts] 8 years
Not had a chance yet Steven.
But i would appreciate your ideas if i have to use the images in the first post.
Thanks
Anonymous 8 years
Patrik,

Would a quick circle radius check not work on differentiating the above images? We tried the follow robofile which seems to work and generates circle radius that are different for each image. Would this work?

STeven.
program.robo
patrick  [14 posts] 8 years
Thanks Steve that works great to differentiate most products by hole size.
But i do have a few products (about 5) which are exactly the same except for the pattern around the hole.
It is these that I think I will need to do an image match.
The lighting should be consistent for each particular product so hopefully the image match will cope.
Pat
Anonymous 8 years
Patrick,

That's worth a try. Another thought would be to expand the circle radius after detected (using the Dilate module) and then mask that against the original image. This would provide just the texture around the hole. This would help the image matching focus on just that area.

Or you may even be able to run the Moment Stats module on that texture to give you a number that does differentiate it against the others.

If you almost get the Image Matching to work post it here and we will test to see what more tweaks we can do to that module to make things work better. It is a very simplistic module and we've not documented it since we expect to improve the functioning somewhat given a good test example.

STeven.

This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index