Embedded Camera

[levlandau2]

This is an interesting concept. I work in a wafer-fab and we have a ton of industrial equipment which uses high-resolution monochrome cameras to do pattern-recs and find fiducials. ~3µm tolerances can be achieved using this alignment method.

[Digitalmagic]

Not sure how to implement it, but I've seen cheap laser pointers (with x-hair pattern) used with a camera for not only automated homing but also range finding to get a touch-less Z offset.

Perhaps the Processing + cheap/good CCD camera + laser pointer combo often used to make 3d scanners has been or could be applied to help a CNC machine correctly reference its workpiece for initial start up then after milling (and debris removed) could scan for conformity and apply any divergent measurement toward corrective machining such as requesting a re-pass over the current work piece or even (though really ambitious) could apply distortion measurements to an auto-compensation routine for future avoidance.

A machine that could learn its own limitations (and work around them) could be really handy in helping an amateur machinist to avoid costly mistakes but unless somebody has a an AI specialist in their back pocket I'm at a loss as to how to achieve it. But it would be SO awesome to see it done.

There is a 99$ LIDAR (Laser+cam distance detection) Kickstarter project still open.
http://www.kickstarter.com/projects/110 ... ?ref=users

EDIT: Perhaps a cheap and lightweight laser distance measurer could be hacked.

[jluther]

Micro machining centers sometimes use high magnification microscopes to track the tool. They locate the part initially then move the machine until the tool is taking off enough material. It's a matter of getting the camera's aligned for each axis (a camera on the y/z plane can determine the y/z coordinates of the tool including runout, then a camera on the x/z plane can determine the x/z coordinates of the tool) at that point it's just image analysis software and lighting to see when the shiny tool gets to the right spot and to keep it there. Telescope guide software might be a good open-source start to look at the analysis of moving to keep an object at the same point relative to the camera.

Problems come in with the angles when the tool is far from the camera, which when machining ten-thousandths is typically not a concern.

Z measurements with respect to the camera would probably be easier to rangefind with ultrasonics than try and accurately measure the size of a dot on the platform or axis, measuring area for that technique seems to work ok though. (laser crosses would get bigger as distance increases and so make the equation more complicated, some scanners use dots of known size on the part so that the distance is known and then the cross size is known, machining with a net of dots doesn't work though.)

2 ultrasonic sensors also can generate multi-dimensional echos.

[Itsjesseyo]

Well Ed, it wouldn't be sooner, but since you are using Linux anyway, I thought I'd pass on this raspberry pi camera module : http://www.newark.com/jsp/search/produc ... ku=69W0689

It patches right into the board, and they've got a python library, which would compliment a computer vision library nicely.

Not to mention its high def and high framerate.i've got two on the way, and i think we have the same kinds of things n mind. I'm been itching to put the computer vision stuff I used to do for artsy projects to use, and I finished the google self driving car class a while back and thought - well damn, I can make any computer system adaptive and smart with this stuff. I wonder how accurate I can make my shapeoko.

[cvoinescu]

Ultrasound that propagates through the air doesn't have enough resolution for this. The wavelength at common frequencies like 24 or 48 KHz is way too long to give even close to enough resolution. The kind of ultrasound used for medical imaging is better (MHz range), but it doesn't couple enough between solid surfaces (transducer face) and air to get any sort of useful image.

Many light-based rangefinders work by measuring angles, not apparent sizes of known objects. The same principle works with two cameras and a target pattern (like human binocular vision), with a light source making a pattern (cross, line) and a camera (or other detector), or with one camera and two targets on a relatively large, rigid object (e.g. Wii remote).

[scott216]

Not sure how to implement it, but I've seen cheap laser pointers (with x-hair pattern) used with a camera for not only automated homing but also range finding to get a touch-less Z offset.

How does this work?

[Brian]

Have you seen this guy:
http://www.cnc4pc.com/Store/osc/product ... cts_id=329

[Nigel K Tolley]

Save yourself some money and just find one of those (fairly distinctive) black bodied webcams for under £10 then strip off the clamp and mod the little stump it clamps to.

If you grind the ball off you are left with a decent approx 1/4" shaft for mounting.

Wind the focus lense out to get a close up zoom for this sort of thing.

Works great. Just not under Linux, which refused to see any of the cameras I bought. Few years ago, so might now be OK, but I have up on it. Work fine under Windows though.

Try and get the metal bodied version, it is slightly better than the particular body version. They look near identical though.

Sent from my SM-N9005 using Tapatalk

[BLAHBLAHBLAH]

Save yourself some money and just find one of those (fairly distinctive) black bodied webcams for under £10 then strip off the clamp and mod the little stump it clamps to.

If you grind the ball off you are left with a decent approx 1/4" shaft for mounting.

Wind the focus lense out to get a close up zoom for this sort of thing.

Works great. Just not under Linux, which refused to see any of the cameras I bought. Few years ago, so might now be OK, but I have up on it. Work fine under Windows though.

Try and get the metal bodied version, it is slightly better than the particular body version. They look near identical though.

Sent from my SM-N9005 using Tapatalk

I'm lost, What camera would that be?