Ask Your Question
0

Detect presence/absence of a screw - Image comparison

asked 2016-04-01 08:45:52 -0500

Hello community,

since several weeks I'm sitting on a problem of object verifications and cannot get sufficiently good results. The task is to check whether a screw is in a hole like seen here:

Image with screw

Or not:

image description

I'm using OpenCV 3.0 through Java and the method I have tried a lot is Feature-Matching through an ORB detector and comparing the number of successful matches in the area where a screw is supposed to be. However, the result is overall fairly unreliable and fluctuates a lot from an image to another (taken live with a camera) even if there are just seconds between the shots and the image is blurred slightly against the noise.

Additionally, the system doesn't seem to get any grip of the actual shape of screw. The features are too small...

Does someone have an idea what other method I could use?

Huge thanks in advance :)

edit retag flag offensive close merge delete

Comments

2

hey, that's a real tough one ;)

berak gravatar imageberak ( 2016-04-01 11:30:00 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2016-04-01 17:35:37 -0500

Tetragramm gravatar image

Oof, tough problem. This may not work very well, but I have some options.

First, is it possible to shine a light through the back of the hole? That would make things very easy.

Second, the screw has the small white reflection in the bottom of the slot. This will only show up on some screws, so it's not reliable. But if you can cause it to be seen consistently, it's a good feature.

Third, the one with no screw has a yellowish tint to it. You could use that to classify. Take the average color and if it's on the yellow side, it's empty. This may need to change based on what's behind the screw.

You can also combine two and three.

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2016-04-01 08:45:52 -0500

Seen: 477 times

Last updated: Apr 01 '16