finding screws
Roland from South Africa  [107 posts]
10 year

Can anyone explain how to best go about using image match?

In the attached image, I took a snapshot of the screw and used that as the shape to match. Unfortunately RR found a match in the open area as shown. I thought I would get a 100% match of the screw since it's a perfect overlay.

How could I better sample the screw image to get a match?
How does scale of the training image affect the match?
Does it help to have a 1:1 scale of sample and target?
Should the training image be highly stylised?


Roland from South Africa  [107 posts] 10 year
It looks like the best approach is to use 'Object recognition' and not Shape match, nor Image match.

What I'm doing is to take a number of pictures with the screw slightly rotated in each one, mostly because of the reflections within the screw.  This way I can get over 90% confidence.

Steven Gentner from United States  [1446 posts] 10 year

I think you figured this one out ... just to summerize:

Image Match - looks at the ENTIRE image as a match. It does not look at particular shapes or features. It instead focuses on color distribution. Works best on scenic images and not on individual parts.

Shape Match - expects a black and white only image (binarized) so a color one will not work. You need to threshold, etc. before using that module.

Object Recognition - more of what you were looking for. Has various techniques that mainly focus on features (corners) or edges. Best for the detection of known parts in an image.

Naturally, lighting will affect the match since the models need to be pixel similar with what you are matching against.

Again, not sure if you've tried the circle or ellipse modules for screw detection. Seems like that would be the way to go instead of actual object/pixel matching.


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