Has anyone used IMAQ to recognize the intersection of the lines in a laser SMR?

[WNM]

Using the IMAQ Vision library, I need to come up with a way of identifying the intersection point of the three facets in a laser SMR (retroreflector or corner cube).  I’ve attached a pair of images of one that I have here.  Has anyone done this previously?

https://www.google.com/search?q=smr+retro

Thanks!

[1100fps]

I have never used IMAQ to recognize the intersection of lines in an SMR.  Never had a need!  Glad for it.

The images are terrible.  If I had to work with images that look like damaged or dirty corner cubes, I would end up writing an edge finding convolution that rotated, to find the lines, then calculate the intersection.  Time consuming and tedious.

Alternately blob analysis might be able to find the center of the cube, but it's hard to guess the accuracy.

I hope someone comes back with a better idea.. I am curious.

I hate writing custom convolutions.

Cheers

Robert

[WNM]

Since I had originally posed the question, our optical engineer got back to me with a corner cube (SMR) image made on the target system using the real system optics and laser light. I've attached it and it is truly terrible. (The first images that I posted were made with the same type of SMR and  camera using a C-mount lens and room light).  After seeing the real image I decided that there was no way to build a reasonable image processing algorithm to find the intersection of the lines. But since a human could still discern them, we have switched to "Plan B" which is to have the human draw two or three ROI lines on the image to define the intersection. That works well/easy enough.  In the case of two lines, there is no ambiguity about their crossing point but in the case of three lines, there almost always will be. In the case of multiple intersections, the software requires that they all must be no more distant then a certain number of pixels from each other before the point is declared to have been found and the average of the intersections is returned as the center point..  

[1100fps]

I'm not going to argue with your solution.

Another way would be to use Deep Learning..  It would require hundreds (if not thousands) of images where the lines were classified by an expert..  It's painful and expensive.

Yep.. Your solution is good.

🙂