2D Scanning for large forms

[Job314]

Hello to all,
I am a beginner in regard to this technology and I need your help for some tips on how to get started .
I have a cnc milling cutter programmable language iso ( g -code) .
I create shapes on the wood, but simply in 2d .
For example, ovals , rectangles, various irregular shapes with lunge z of about 5-6 mm .
My problem is that I often forms huge , kind of 1-2 meters and manually recreate it becomes a waste of time incredible.
My goal is to scan these objects or drawings, putting it on the work surface of my cnc milling cutter , get the drawing and convert it to ISO G -CODE .
To do this, I would have thought to use the method of david, mounting the camera and the laser, on the head of cnc, creating a program in iso that makes move around in a motorized along the x and y axes where i put my form to scan.
As I thought about laser use an inexpensive red ( the cheapest kind of shop of david )
It's possible? And above is correct as a system ?
My questions are :
- What webcam do you recommend to use?
- The laser is good or is economically better one with adjustable focus ?
- How can I calibrate the webcam so that gives me the correct distances ? Using a calibration panel PLANE instead of creating the angle for the 3d ?

Thank you all for the help

[MagWeb]

I removed, as suggested; your post at SL scanning.

If I got you right, you wish to digitize simple 2D shapes at your CNC? Not sure if DAVID is the right tool to do this. Did you consider a touch probe sensor on your CNC?

DAVID is great to capture 3D shapes as carvings and similar stuff and there are ways to do this mounted on a CNC. Different to a CAD drawing DAVID describes a shape by connected points. This means a simple square isn't described by only 4 connected points but by maybe thousands of points (somehow the same difference as vector or pixel based drawings)... This results in a finally not that accurate toolpath...

DAVID can only calibrate on a 3D object (the corner) not on a single plane. If you are trying to digitize on your CNC (with a limited height in Z) you'd need to to scan your big sized object in several parts and mount these parts to a single surface in Shapefusion later on.

[fred_dot_u]

My opinion only, based on your request, considering large area of scanning to be done.

Instead of using DAVID on x-y of machine, use ordinary digital non-video camera. Set z to maximum for most area in camera view. Measure first time for x distance, y distance. Use cnc g-code to move same distance in x full area, move y distance, then full x distance again. Each time you stop, take photograph. Use stitching software to combine images into one large image. It may be necessary to add calibration marks over surface of item being scanned to assist stitching software, unique markers to match one photo to next. Think of chess board with camera taking photograph of each square, then combining to make chessboard in one photo. For your project, the chessboard is not square at edges and does not have color changes for each camera location, which is why unique markers may be needed. Even dots scattered randomly over surface may work

When images are stitched into one, many CAD programs permit tracing or background adding to allow conversion to vector. I think simple programs like SketchUp allow tracing. Also some software for raster to vector conversion may perform conversion automatically. If item being scanned has high contrast to background, the result could be done easily.

If you wish to use DAVID, it would be necessary to build corner large enough to support your item to face camera and projector. I think it would have to be not part of the cnc machine for the scanning portion. The result of using the corner would be faster than the above photo/camera idea.

[Job314]

Thanks for the replies .
So you say that the solution of david for my large scans is not suitable as it is structured for 3D.
You advise me to take pictures , put them together and then turn them into vector image ISO to be able to convert to G -CODE .
I was interested in the solution of david especially for accuracy , my fear is that by doing as suggested , the accuracy is not the best.
I state that I do not know software designed for this purpose , and I suppose to manually merge the images with paint is not the ideal solution .
I also have another doubt : I tried to convert to vector scans of relatively small objects (a4 ) by using my scanner and the program of Vectric VCARVE 4 .
Ok for proper operation , however, creates numerous imperfection . For example, an oval shape instead of create me the linear path following the profile of the form , he create points / nodes, that interferes with the regularity of the design. Should I open the drawing , zoom in , manually delete these points / nodes and regularize the form . Practically becomes an absurd work .
I was hoping to find a better solution that would make me all the more uniform.
For this I thought of the solution of david, inserting the laser in the head of cnc, webcam fixed at one point and with the regularity of the movement, I was hoping for a more accurate and consistent.
Any advice on software what i can use for merging images and create vector? Thanks again for the help.

[fred_dot_u]

I think that my suggestion to stitch images together will not provide accuracy for your purposes. The result would be no better and perhaps worse than your work with small items on scanner.

For laser scanning, there is the requirement to have a calibrated corner behind the object. The camera views the laser line on the object and also the laser line on the corner. If you have large object and one large corner, accuracy is limited by camera resolution, so it is not going to be good accuracy. As an example, I picture in my mind a building corner with your object leaning inside and supported above floor a small amount. The camera is placed to see the entire object and part of the two walls of the corner. If the camera is 800 pixel resolution horizontal, accuracy is limited to 800 pixels. As part of this example, consider that your object is 0.75 meters wide and placed in the corner so the camera sees 1.0 meters. In this example, 8 pixels is 10mm or one pixel is 1.25mm and larger items would have lower resolution and lower accuracy.

If using David Structured Light Scanning, you would have to have a calibrated corner, but could then remove the corner and scan smaller sections of your object. If the object is flat with no detail, there would be difficulty to join each small section together to make a full object.

Returning to the camera mounted over the object and using stitching software: It would be useful to have a scale placed on the object if possible. The camera would take an image, including the scale, the stitching software would join the sections and you would be able to see if there is displacement error. If there is no error, you would be able to use the scale on the object to calibrate for accuracy. I have done this on ordinary scanner and the error in joining four pieces was about 1 mm but it was not as large as you suggest.

In place of a scale on the surface of the object, there could be graph paper with calibrated size squares under the object. The camera would record the squares and provide measurement reference in that manner. In the USA, paper with 0.1 inch squares would provide that level of accuracy .It would be challenging to attach a large number of paper together and to keep them square and accurate.

You have a very challenging project in mind!

[VDX]

... some years ago I've made something similar with vectorizing floor tiles in a cathedral with a camera setup - the camera was placed maybe 4 meters above the floor what was good enough to avoid big distortions, so the geometrical accuracy was sufficient.

For 'calibrating' a distorted 2D-image you can first take a picture of a raster with known spacing and calculate the inverse distortion to get it orthogonal again ... then take your image and 'stretch' it with the same values ...

Viktor

[WalterMo]

I think there are two possibilities to laser scan big flat and only a bit (in z direction) structured objects:

1.) By planeless scanning like Mattia had done it here when he scanned a Matisse sculpture:
viewtopic.php?f=7&t=3223&p=17886&hilit

But planeless scanning to do wih a laser is rather elaborate regarding the laser sweep mechanism (electronic and mechanic).

2.) Easier is to try it with my old flaps corner:
viewtopic.php?f=3&t=133&p=597&hilit

and here (April 28th, 2010):
viewtopic.php?f=1&t=1825&p=11766&hilit

Laser, corner and camera must be mounted at the CNC machine. The laser must have a separate sweep mechanism. After calibration with closed flaps the unit of opened corner, camera and sweeping laser will scan the object in several segments. These segments will be stitched later on by DAVID's Shape fusion.

PS. At that time when I scanned the tread of a car tire I had done the laser sweeping by hand.

[Job314]

Thanks to everyone for their valuable assistance.
These days they are a bit 'busy due to urgent work to be delivered. As soon as I can I will get rid of all the evidence in the case, following your advice.
I will keep you posted! Thank you again!

[Sunderz]

I haven't had chance to read all the posts, but I use a large cnc mill at work also 2600mm x 1460 mm. I use cam bam on some of my projects and that can take in the g-code and reverse engineer a part from the tool path.