Reed Knot Recognition
Hello everyone =)
I'm a Bachelor student from Switzerland in mechancal engineering and I'm looking for a solution in a project. I have to say, that I have no experience in OpenCV and I am only looking for an answer if OpenCV would be worth learning for that application.
So let's get to my problem: In the project I have to build a machine, that cuts reeds in a specified length and positions the knot in a specific position (normally the end). To position the knot I somehow have to recognise it and have my controller be able work with a position of the knot.
Here's a link to some pictures of (already cut) reed pieces to make you clear what I'm talking about. https://drive.google.com/open?id=0B20qWrfA2pxpR3hUVlRnaHl5c28
Now my question is: Would OpenCV be able to recognize these knots with a high reliability while the reed is moving with ~0.2 - 0.4 m/s,
Thanks in advance for your effort
Lukas
Seems to me like an application OpenCV can be used to help solve. Do all of these 'knots' tend to have a different brightness feature passing around the circumference and potentially a dip in the radius there with slightly different radii on either side? The reliability of the approach depends quality of image data coming into OpenCV (includes illumination and optics for clear high-contrast imaging, and to some degree quality of the material). Having a background pattern or stereo depth imaging could help quite a bit too.
Thanks for your answer.
In the same Drive I uploaded some pictures of reeds, which should clarify the question about the difference and range of brightness of reeds and knots. As you can see on the pictures every knot looks a little bit different from the other (but overall they are all similar). I don't know how OpenCV can handle these differences. I think a contrast between knot and hollow reed should be achievable with a good background and illumination in most of the cases.
Whole other question: Would OpenCV also be able to determine the diameter of the reed just by the image? Because an appreciated feature would be that reeds with a diameter above a certain threshold don't even get cut and so don't have to be sorted out manually afterwards.
I feel that you could write an application to apply OpenCV to help sove these problems. The lighting and placement of items in the field of view looks pretty good. There's usually lots of filtering one needs to smooth the data and model the observed data as items (reeds), and that's part of your application code, with some help from OpenCV primitives. You'll be able to measure the diameter of the reed, at least in the plane of the image. To get height (depth from the imager), you'd need some depth measuring system (stereo disparity, optical flow, a line projected across the belt, a light curtain, or something similar). This can help with accuracy of diameter measurement, but if all you need is an approximation, you wouldn't really the added complexity of depth measurement.