Scanner Data Samples

[Scanner Man]

Object color, texture variations, highly reflective surfaces, ambient lighting and translucency to name a few. White light technology does not suffer from these same issues.

Red laser systems - Most are triangulation with 1 or 2 cameras, Some use interferometric techniques which have completely different strengths and weaknesses than the triangulation systems.

White light projection systems do have issues with color, translucency and highly reflective surfaces.


Please note that all the systems do not suffer to the same degree and allot of factors can affect if a system can measure something. So the following is a general rule but you can always find exceptions where they will work.

Laser systems that use a camera will suffer from:
Ambient light
Object Color
Highly reflective materials
translucency

Interferometric techniques suffer from:
Highly reflective materials
surface texture
translucency

White Light systems suffer from:
Object Color - (Try and measure Impregum on a white light system)
translucency
highly reflective materials

 

[NicelyMKV]

Interesting. So far black is the only color my white light system can not pick up. I do notice quite a bit more detail in the scans compared to red laser though. Care to tell us which scanner you are involved with? Very interested

Just realized I am hijacking the crap out of this thread and going waaaay off topic. I will start another thread scanner man if you would like to talk about the system you are using. Sorry guys.

Jason

 

[Scanner Man]

"If I aim my scope at the bulls eye, I am confident that the bullet will hit the center of the bulls eye +/- 2inches 75% of the time and +/- 5 inches 90% of the time and +/- 8 inches 99.99% of the time"

For a scanner to claim "20 micron accuracy," it means that for every single data point in the mesh, it could be where it says it is +/- 20 microns.

The dental industry is not really defined in terms of specifications and at least some standard tests will arrive when ISO 12836 is put into practice but the game of specs will still exist after implementation.

If a company specs the system to 5um's. What does this mean...
They are really saying +/- 5um's at some sigma. Is it 1, 2 or 3 sigma.
At 1 sigma only 68.27% "of the measurements" falls within +/-5um's , 2 sigma - 95.45% and 3 sigma - 99.73%

What is "of the measurements"? Some companies define it as the accuracy to a standardized object after the STL.
For example if a 3mm diameter ball is scanned and the resulting STL dia is 3mm dia. Was the result an egg shape that had a dia of 3mm? Since no Form value or std dev. is provided this method is meaningless and large scanning deviations will hide within the standardized object result.

Other companies use the % of the STL surface area that falls within the tolerance.

So what is 20um accuracy? I can definitely tell you that it does not mean that every single data point will fall within +/-20um's and I can guarantee that some % will be higher than the spec.

The data points in the resulting STL are never the raw data points they have been filtered, messaged and enhanced to provide a more accurate result.

In ISO 12836 Trueness is defined as "Calculate the difference between the mean of the 30 repeatability measurements and the true value."
The good thing about ISO 12836 is companies must also provide the std deviation for all measurements in their report.

 

[pepsi185]

The dental industry is not really defined in terms of specifications and at least some standard tests will arrive when ISO 12836 is put into practice but the game of specs will still exist after implementation.

If a company specs the system to 5um's. What does this mean...
They are really saying +/- 5um's at some sigma. Is it 1, 2 or 3 sigma.
At 1 sigma only 68.27% "of the measurements" falls within +/-5um's , 2 sigma - 95.45% and 3 sigma - 99.73%

What is "of the measurements"? Some companies define it as the accuracy to a standardized object after the STL.
For example if a 3mm diameter ball is scanned and the resulting STL dia is 3mm dia. Was the result an egg shape that had a dia of 3mm? Since no Form value or std dev. is provided this method is meaningless and large scanning deviations will hide within the standardized object result.

Other companies use the % of the STL surface area that falls within the tolerance.

So what is 20um accuracy? I can definitely tell you that it does not mean that every single data point will fall within +/-20um's and I can guarantee that some % will be higher than the spec.

The data points in the resulting STL are never the raw data points they have been filtered, messaged and enhanced to provide a more accurate result.

In ISO 12836 Trueness is defined as "Calculate the difference between the mean of the 30 repeatability measurements and the true value."
The good thing about ISO 12836 is companies must also provide the std deviation for all measurements in their report.

I am really impressived by your knowledge in scanner, can't help replying.

 

[patmo141]

If a company specs the system to 5um's. What does this mean...
They are really saying +/- 5um's at some sigma. Is it 1, 2 or 3 sigma.
At 1 sigma only 68.27% "of the measurements" falls within +/-5um's , 2 sigma - 95.45% and 3 sigma - 99.73%

Assuming pure Gaussian distribution. I just gave an off the top of my head distribution to demonstrate that an uncertainty should really be a distribution of probability and that accuracy of _____ microns is not very descriptive

What is "of the measurements"?

I think this is the hardest question to wrap ones head around or define clearly. In the point cloud of a SL scanner, each pixel measures a depth (well, an intensity which maps to a depth right?) but this measurement is repeated over multiple SL patterns, and then processed...and then merged with another iteration from perhaps a different view point, and then processed by one of several "meshing" algorithms. So, at what step along the way does our "measurement accuracy" get assessed or reported? At the point cloud? At the final meshed surface? I can see merit in both

Other companies use the % of the STL surface area that falls within the tolerance.

This seems logical enough despite the weakness of "shape specific weaknesses hiding in the overall result". How about, in addition to "averages," std deviations etc, a colored mapping of the Hausdorrf distance would show where the surfaces is most faithful and where it is not. That way, as you stated, a [scanner + all algorithms] can be assesed for strenghts and weaknesses across a sample of objects.

So your egg would show error at the poles.

***'s scanner might show error on sharp edges whose crease points perpendicular to scan direciton (marigns),but great accuracy on flat surfaces whose normal is parallel to the scan direction.

The data points in the resulting STL are never the raw data points they have been filtered, messaged and enhanced to provide a more accurate result.

This is something I have wanted to get across for a while. The mere process of meshing a point cloud is a processing step, meaning it's not "raw data"

 

[Scanner Man]

I think this is the hardest question to wrap ones head around or define clearly. In the point cloud of a SL scanner, each pixel measures a depth (well, an intensity which maps to a depth right?) but this measurement is repeated over multiple SL patterns, and then processed...and then merged with another iteration from perhaps a different view point, and then processed by one of several "meshing" algorithms. So, at what step along the way does our "measurement accuracy" get assessed or reported? At the point cloud? At the final meshed surface? I can see merit in both

Your description of the process is correct. More detail goes into determining what points to trash, keep and/or average with surrounding points.
The reported accuracy from scanner manufacturers is at the final STL output.
The point cloud accuracy is never reported.

This seems logical enough despite the weakness of "shape specific weaknesses hiding in the overall result". How about, in addition to "averages," std deviations etc, a colored mapping of the Hausdorrf distance would show where the surfaces is most faithful and where it is not. That way, as you stated, a [scanner + all algorithms] can be assesed for strenghts and weaknesses across a sample of objects.

To really access strengths and weaknesses you are correct. Internally all these tests are done on a large variety of difficult and easy shapes.

So your egg would show error at the poles.

You are correct and that is why I said the Form error or Std Deviation is also needed or else this type of measurement is meaningless.

This is something I have wanted to get across for a while. The mere process of meshing a point cloud is a processing step, meaning it's not "raw data"

A pure straight forward meshing would not change the raw data but NO one does a pure mesh.

Every system changes the data to merge multiple images, reduce noise, smooth surfaces, enhance highly curved features, fill holes, trim edges, delete unwanted points, delete redundant points, delete intersecting triangles, delete tunnels, subdivide polygons and decimate.
All this is done to provide a error free watertight stl for the CAD.

 

[DMC]

Pat,

You can re-position the dies to get better angle of margin area and re-scan if you really feel the need to capture more data.

I had a wee bit of experience years ago playing with a NextEngine scanner.

IT was fun to have full control of the scanning process, even with a little $3000 laser box. Great waste of tme on the weekends!

You could set up the entire scanning sequence, number of scans at ? angle, etc..

Then I would play around with each individual scan before I reduced noise, trimed away rough edges at fringe area, aligned and merged, etc....

 

[DMC]

I put a coffee cup over the camera lens with a hole cut out of bottom.

This is Red-Neck way to reduce point cloud and to speed up processing of data for single die. LOL



Sprayed my son with scanspray years ago....
I could not get him to sit still long enough to get the Seven shots I wanted. Oh well.