Re : Re: Measuring a rotation speed

[muks]

---

Bruce Ammons wrote:

Here is what I would probably do:

Train a pattern for the N.  You might want to try standard pattern matching or geometric pattern matching and see which one works better.

For each image coming in, use pattern matching to locate the N.  Make sure you allow full rotation.  Once you determine the center of rotation, you should be able to measure the angle between the N and the center of rotation.  Each frame should give you a delta angle from the previous frame.  The formula for instantaneous speed will be: (delta/360 deg)*frame rate = rotations per second.  This won't be a really smooth value, but it should be very close to true speed.  You could filter it if you want a more average speed.

Bruce

---

Bruce did you see this?

---

The motor axis (and of course the mechanical piece) can rotate from 0 to 10 rot/sec.

---

 do you think acquiring and pattern matching can match up to that speed? 

[BruceAmmons]

If you know it always rotates in the same direction, 15 fps would be adequate to measure the rotation speed.  A small rotation in the reverse direction would actually be a large rotation in the forward direction.

I'm not sure how long pattern matching would take.  If you have a fast computer and tweak the settings a little bit, you could probably find the letter N at 15 fps or faster.  I would need real images to test and see how fast it can actually go.  You wouldn't use subpixel resolution or other options that slow it down considerably.

If the N has high contrast with the background, it might be possible to threshold the image and search using binary tools.  They would be much faster than pattern matching, but require high contrast.

If this doesn't work, I have a few other ideas that might work.  This is the simplest one, though.  I don't want to confuse things by suggesting other options right now.

Bruce

Message Edited by Bruce Ammons on 01-15-2010 11:47 AM