OpenCV vision with EMC/linuxcnc

Ok guys, you win the cameras on mill/lathes.

But the PnP market is still there.

YEs!

Have touch probe for milling, would still like to work on pick and place vision - where do I start as an experienced vision systems coder but EMC source novice? I know I've got to do a lot of reading, but some practical advice would be great

Cheers,
Ed

I'm slowly working my way through building example routines for all the core needs, detecting part presence, comparing to reference, detecting offsets as you mentuioned etc, but I'm keen to ensure that my approach will fit together with EMC without too much awful spaghetti code.

I'm sure I can do the vision, how much others are willing to help in this and other areas will largely decide the feasibilty/outcome. I'm sure given enough time I'd make my way to a working solution - it would be great to arrive at a sensible solution for a system including the definition of the pick locations etc.

Working from gEDA's PCB I get centroid and rotation data for PNP, from this I have a good idea of the control that's needed to make placement happen, it would be great if someone could work in tandem to decide on an approach for defining the pickup locations of the parts, reel dimensions and offsets etc.

Thanks!
Ed

Regarding keeping cameras clean in harsh environments, I have seen (but not used) a pinhole camera system that uses compressed air to keep the lens pinhole open and clean. These are supposedly great in placed like paper mills for viewing the sloppy bits of processes and would more than likely fair quite well in a lathe/mill environment if mounted in the right spot (and you weren't too concerned about the need to supply 20cfm air to a camera!)

Just a thought

I am a hobbyist with my first CNC machine (a Probotix V90 mill.) I looked at using a webcam instead of limit switches, since the webcam would be cheaper and easier. Note that on this class of machine. exceeding the limits is merely embarrassing, not damaging.
(except for -Z, but that's another problem altogether.)

For this use, I would place the camera in a fixed position on the enclosure, behind the gantry, and use it to locate a square painted on the back of a piece of metal mounted to the Z axis. Accuracy need not be better than 0.1" and I suspect that just about any machine vision system, with just about any camera, could handle this single, trivial, completely reproducable problem.

Cost: limit switches are "only" about $2.00 apiece: you need 5 of them, plus you need to run the wires, plus you burn I/O pins. A cheap webcam costs about $20 except that I have several in my junk box, and you need only one fixed USB wire, not 5 moving wires.

A Webcam that can get you within .1" can be used in conjunction with a touchplate. Use the webcam to get close using rapids, then touch off for final accuracy.

I have not pursued this, since I have lots of other machine improvement projects in play. (I added the limit switches, but I would still prefer to use the camera.)

Are you saying you are not using the swtches for homing?

The single biggest help anyone could give me right now would be a pointer as to what to read first in the source code, even better would be a template/starting point/some instruction for creating HAL objects.

I'm happy once I've got myself a 'framework' that will sit well with EMC, if such a thing is available it would be great - alternatively I'll publish something that does when we arrive there.

Cheers!

Ed