Defect detection

OpenCV, Python and other image processing questions
Post Reply
Posts: 1
Joined: Sun Aug 30, 2020 5:45 am

Defect detection

Post by Amey18 »

Hello everyone,
I am building one application for defect detection in some plastic molded parts. There is a circular part where generally defect occurs. Can anyone please tell me how to catch the defect? Attaching some good as well as defected images. Thanks in advance.
IMG_20200829_134853.jpg (1.6 MiB) Viewed 398 times
IMG_20200829_134941.jpg (1.57 MiB) Viewed 398 times
IMG_20200829_134940.jpg (1.55 MiB) Viewed 398 times
IMG_20200829_134959.jpg (1.65 MiB) Viewed 398 times
IMG_20200829_134959_1.jpg (1.54 MiB) Viewed 398 times

Posts: 189
Joined: Tue May 21, 2019 12:33 pm
Location: France

Re: Defect detection

Post by stereomaton »

Your problem does not look related to the stereopi, so you should probably ask the question on another forum.

What I personally could say is that you better have to work in a controlled light environment. In particular, you can use the light to illuminate the zone to inspect only. With similar angle as in your samples, you can for example try to detect the shape of the reflection on the top. Not sure, but the simple blob detector of opencv may help. Again specialized vision forums would give more pertinent answers with possibly more robust ways to deal with the problem.
Stereophotographer and hacker
Despite my quite active participation in the forum, I am not in the StereoPi team
StereoPi Standard Edition + CM3Lite module + a few cameras

User avatar
Site Admin
Posts: 603
Joined: Tue Apr 16, 2019 9:23 am

Re: Defect detection

Post by Realizator »

Amey18, one of my friends is developing a similar system now. They need to check the quality of the plastic components (size and form). For this approach, they are using OpenCV to build a point cloud from the depth map (using a stereoscopic photo from the StereoPi). And using this point cloud, they are doing all measurements.
But this method has one specific. It is difficult to build a good depth map, if your surface is either glossy or not textured.

They are using a couple of HQ cameras with the 1920x1080 resolution for the depth map building.
As Stereomaton says, for your case you can try to use a single-camera solution and another set of techniques to detect defects (like NN, or OpenCV algorithms).
Eugene a.k.a. Realizator

Post Reply