Vision System: Registration
 

This thread is to pursue R&D around the concept of Vision System R&D on the MechMate CNC router.

Initial topics requiring research:
- Vision software
-- MatLAB - expensive but very capable. A professional tool.
-- OpenCV - opensource, very capable, extensive libraries. requires a steep learning curve and significant development skills to become proficient. may require specific libraries to be custom developed
-- More
- CCTV
-- IP cameras - there is a wide range of IP cameras available for very low prices. The camera probably doesn't require its own gantry. I would put it in the spare payload position on the existing gantry.
- Integration with Mach3 or LinuxCNC
-- Mach3
-- LinuxCNC
- Targets
-- use material itself as target. Vision system can detect edges and depths and other features. I'm not sure what accuracy or resolution.
-- use specific orientation targets. stickers, pen marks, painted corners. This solution is the fallback, it is better if solutions require no specific targets but can orient on material.
-- nested orientation. Is it a requirement that vision system can detect just sheet material edges and orient dimensions or is it a requirement to detect sheet material internal use. ie. Does the vision system system need to analyse a whole sheet and detect previous cuts to find a free space to place the new cut.

Keep the ideas flowing. Add either your wish list or your skills and willingness to contribute to this offshoot project to MechMate.

I've just finished building an extended table (11' 6"). A while back i acquired an inexpensive USB microscope. To check the end to end wander I pulled some 20lb fishing leader taught, mounted the USB camera on the gantry, focused at 200x on the leader and recorded a video as i pushed the gantry down the full length. After that I used ffmpeg to create a series of jpgs which i could then inspect to determine the wander. According to the pixel count, I have a bow of less than 0.004 over the full length. At this point, that's good enough for the cutting i'm doing.

Wow, that's awesome Robin. Surely the bow in your material is going to be much more than 0.004 at least with ply and mdf. Great idea with fishing line too, Ive added it to my list of targets.

I should clarify, the camera was viewing downward so the bow is side to side not vertical.
To measure the vertical variance I would have to take into account the line sag. eg on my table the sag would be about 0.005 at the middle of the catenary. I would likely have to surface the table and use precision gauge blocks to offset the line above the table in order to take the images (readings). Top down was much easier to do as a proof of concept. :-)

cheers

a bit off topic but: ..... before you would measure it top to bottom you would first need to make sure your X-rails are perfectly straight and parallel in all fields, to be able to know what you are measuring is indeed the bow in your table ... and not something else. And then the fishing line and microscope is better replaced by a digital micrometer on a magnetic base in the middle of your gantry, which you zero on the deepest or highest spot and travel over the table.

I suppose I should be more careful with my words. What I am measuring is the wander in the the rails rather than the table itself. The same applies for the vertical measurement. Once you know the variance in your rails (X & Y) your table after surfacing will follow.

And, for the $20 that the usb microscope cost it was a fun little experiment :)

cheers

Ah I see, you are then measuring the wander between the two rails imo not each individually. And the surfacing indeed makes the table top follow the straightness of your machine .....but all is relative.
See my thread about preload shimming the rails on the gantry on here.

Check out http://www.machsupport.com/forum/ind...c,22932.0.html for a great quick and dirty way to use vision for registration. It's a manual position method that uses three points to locate the work.