Possible with OpenMv? Detecting difference and surface flaws

Hi All,

I am new to OpenMV and deciding to buy OpenMV Cam M7, so I hope you’ll bare with me. I have an automated system applying a red glue with thickness of about 1.5mm on the edge of a plastic mold. Sometimes there are flaws in the gluing, I mean non-continuous dispensing and bubbles. I attach a picture of a good vrs flawed gluing. Can I use OpenMV Cam M7 to detect these flaws and flag the gluing as unsuccessful? For example look for a broken glue line or an air bubble larger than 1 mm? I don’t mind getting my hands dirty with coding. Hope you guys with experience can tell if this is at all possible to do. Thanks in advance ladies and gentlemen! :smiley:

Yes, we support frame differencing and then you can check if you see differences in certain ROIs in the image. Assuming all pieces in the system are fixed and everything lines up always this is very easy to get working well with the OpenMV Cam. That said, it works only if the camera is in a fixed location, you have good lighting, and the prices of plastic are in a known location too.

As for the ROIs, you’d create a lot of them such for where the red stuff is. Alternatively, one big ROI could also be used. But, then you won’t be able to tell where the problem was exactly.

Hi Nyamekye,

Thanks for the quick reply… really appreciated. Yes I plan for the camera to be at a fixed location and the products will be always aligned to an accuracy of about 0.3mm. I plan to have the product in an enclosure with controlled lighting also.

Since the glue is fluid, there will be variations in the amount (streak thickness etc)… I am more interested in detecting lack of continuity in the red stuff applied. If I can slice up the regions of interests, can I detect lack of continuity in the ROI? What tools, or libraries can you suggest for doing this. Thanks a lot! I’m in the process of ordering a couple of boards.


The OpenMV Cam has all the functionality you need on-board without installing libraries. The whole point of the product is to have all the computer vision stuff in one place versus dependency hell.

Anyway, this method does the trick:


You can set the thresholds to select for red. Then, you can setup rectangular ROIs all around where the red should be.

Hi Nyamekye,

Excellent! Thank you very much for taking time to explain things to me. Much appreciation.