3D Scanning

OK this is one carried over from the Mendel Tool Head Wishlist.

To kick off

How about a distance measuring pen that can be put into Darwin (or anyother) in place of the extruder.

3D Scanning could then be literaly a scan of the object on the bed (not used during the scan) and will give a topo plan.

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Just to avoid reinventing the anti-grav unit--

[www.muellerr.ch]

and

[www.volunteerlabrat.com]

obviously these are not integrated into the toolhead/table combo, but that is not a problem for me as I see the RepRap as part of a larger self reproducing workshop.

I see

This is similar to aome technology A company I worked with used to measure the clearence around rail tracks from an instrument car.

(I actually saw the unit again just recently at the train station in Doncaster so I guess it is still on the go)

they used a very bright light in the center of the instrument car directed with a pair of flanges around the skin of the car and had CCTV gear mounted at each end of the car pointing towards the central flange/light source.

The Band of light projected onto the area surrounding the car was picked up and it's distance from the rail car was calculated by trig (it was a right angle triangle type calculation) each vertical line of pixels was in effect illuminated by reflection from something a known distance from the side of the rail car.

I shared an office with the poor guy who had to go through the video looking for where the rail track had shifted too close to something and needed moving.

I was writing Transputer code in Parallel C at the time for doing ultrasonic NDT testing of the rail track at 40Mph.

Fun stuff.

The main thing is the principle.

I am struggling a little at the moment as to quite how to adapt this to the capabilities of the current Darwin unit.

Some of the turntable based stuff that has been experimented with though has to be a dead ringer.

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Also lets not forget these

[www.splinescan.co.uk]

[www.vision.caltech.edu]

Wow-
That last one is basically a software only solution. I mean yeah, you need a desk lamp, a checkerboard, 2 pencils and a ruler, but take a fiver to the dollar store and you're done.

Limitations: one side rather than full 360, and probably not so good on dark colored objects but still...

Most interesting.

All seem to exploit the same principle. that of an optical linear reference bounced of an object and its apparent displacment by a 3d object picked up by a camera/array placed at an angle to whatever generates the linear reference. (Linear only in that if projected onto a flat plane it makes a straight line)

Main differences being, linear shadow v linear strip of light (coherent laser light in one example)

The bottom line is that provided the linear optical reference provides a high contrast edge to detect off it does'nt matter what you use (Shadow, Coherent Light, Non Coherent polychromatic light etc)

Interesting.

I guess the color of the object, surface texture and angle of the plane of the incident surface can introduce a degree of error to the result (but maybe not much or does it ??). IE anything that blurs the high contrast edge in any way

It's almost a bit parallax like.

I guess the fact that the light is a point source rather than being a strip source will add a certain degree of distortion too, not to mention the camera and optics. I guess the CCD element layout can introduce some error too.

The essential question being can it be corrected or are the results good enough for the intended purpose though ??

I wonder if polarizing either the light source or image via a polarizing lens could remove some of that error ??

Thoughts, what do you guys think.

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

I think it's time to break out the hardware and run some tests, but I don't have the time, so I'll shut up now.

monoslam should be able to build 3d models of objects if tweaked to give a much higher density of keypoints.

see youtube.com/activevision for demos of what it's capable of

I found this recently, haven't tried running it yet:
[ezra.dev.java.net]

Easy Shape Reconstruction Application in Java

EZRA is a uniquely designed neural network (NN) for reconstructing three-dimensional (3D) object's shape from its two dimensional shaded images taken from multiple views. It is an analytic, dynamic, adaptive NN that can refine an initial polyhedron model to approximate the object's shape given several different view images of the object. Development of this neural network was funded by JSPS (Japan Society for Promotion of Science). in April 2005 - March 2007, as post-doctoral fellowship.

this is a nice, cheap scanner technology that seems to produce some good results... [www.david-laserscanner.com]

Interestingly, the guy I work for now has developed a nifty stereo webcam, which could potentially be used for this sort of job. One of the next projects I may be working on involves isolating a (human) subject from a noisy background, and identifying things like gestures, or whether they are actually looking at the camera.

The API is .net though, and I'm not sure how much of the code is released, nor under what licensing terms. I'll look into it.