Finding a corner with Robotiq Camera

One of my customers asked me if he could locate a corner with the vision system.

This is a tricky application since the system will always process closed shapes and for a corner on a big part, you can't have this closed shape.

What I decided to try is to teach a triangle just like in this picture:

But there was some challenges...

When it was time for the camera to locate the same triangle on a part, 30% of the feature could never be recognized since the 3rd side of the triangle doesn't exist. That being said, the threshold should be less than 60%.

The other challenge is because the triangle shape is simple, it can be recognized easily anywhere where it can find 2 lines intersecting. So any details on the part surface can bring a false positive in the corner detection.

In this example, I kept the triangle taught on the part to avoid the potential false error as mentioned above.

https://youtu.be/kasAWsPClZM

Another interesting detail is the relative position recorded by the vision system are overwritten each time another "cam locate" node is used.
Note that because I used 2 "cam locate", one for each corner, I had to use an assignment to make sure the position weren't lost. This explains the script:"movej(var1);movel(var_2)..." See the program below:

In order to use the corner detection in a production environment, the corners should be identified the same exact way using a sticker with a defined shape. However, this adds a step in the production and can bring other challenges. What do you think?

In the end, I won't recommend to using the vision system for open shapes like in this example. The detection score is too low, it can give false positives and the counter way to do it can bring other challenges.