DAVID 3.9.0 beta1: Reduced SL Distortions

[Sven]

Hello everyone,

We have been working on the projector calibration. In some cases, depending on hardware setup, the scans were a little distorted, mainly at the borders of the scan area, and investigations indicated that the projector calibration was not precise enough for some projector lenses.
We now use a more complex mathematical model and results are better.

I have tested a lot with the hardware I have, and found no problems. But I would like to ask you all to test the new version with YOUR hardware/setup and see if everything works fine, especially if the SL calibration works fine, and all scans are at least as good as before, and in some cases less distorted at the borders.

Download: http://www.david-laserscanner.com/freew ... _beta1.zip
EDIT 2012-07-22: Bug in GERMAN language. German users, please switch it to English.

Especially if you have had problems aligning scans 360°, the new version might help.
Please post any comments or results you may have. THANK YOU

Sven

PS: Other changes, for now, are:

New Features:
- If license not ok, show warning at startup
- Fusion resolution: Show resulting "vertex spacing"

Misc:
- More precise projector calibration
- Fusion: "Resolution" setting is now a slider
- Show tooltips also on disabled menu items
- Contact Pair Selection: Larger mouse pick radius (red circle)
- Saving with texture: Avoid non-ASCII characters in file names
- Screenshot filenames: Hour from 0-23, not just 0-11

Bugfix:
- Remember and try to restore FPS setting of camera
- Avoid degenerated triangles in Fusion result
- a few other small fixes

[Sven]

About Fusion resolution:
You may wonder what the "Vertex Spacing" is. Unfortunately it's difficult to explain what the "Resolution" and the "Vertex Spacing" are.

Regard the Fusion result as a 3D point cloud. It is stored in a 3D cell grid, consisting of up to (Resolution*Resolution*Resolution) cells. The size of each cell is VertexSpacing*VertexSpacing*VertexSpacing mm.
In each cell, there can be only one 3D point. But the 3D point can be located anywhere within its cell!

So the 3D points to not have constant distances, in average it's "VertexSpacing" mm.
The resolution (number of points) is given by "Resolution", but you can't compare that to the regular resolution of e.g. a camera chip. The precision of each 3D point can be much higher than "VertexSpacing"!

I hope this is comprehensible...
Sven

[AlexNoir]

Hello!

I tested it only for aligment and Fusion. Both steps went fine, but during the saving of the fusioned model the software broke down (two times) I'm now using 3.8.1 again.

Alex

[Sven]

Hi Alex,
Thanks for your report!
Sorry, I found the problem. It occurs only in the GERMAN language. Please switch the software to English.
Sven

[ShadowTail]

Here's a scan I did with the beta version:



I used a vertical setup for this.
Hardware: Logitech c920 webcam (1920x1080@30fps), benq joybee gp2 projector (1280x800)

Textures end up being quite blurry as its just the stock hardware without any modifications or special lenses.

If you'd like to see the actual result, you can download the fusion mesh and texture here:
https://dl.dropboxusercontent.com/u/25588164/CandleHolder.rar

Overall the quality and accuracy has improved as previous scans of this object never quite matched up.

[AlexNoir]

Hello!

So I now tested the whole process and it work out fine (okay noww I used the english Version) BUT only one Point. The Software were not able to "fail" the calibration it means that it accept complete irrealistic results (too wide range and angle) - First I was surprised and tested to scan and there was a complete distorted scan. I changed the Version down to 3.8 and there faild the calibration. I canged the view to the Panel and got a result off calibration and run it 3 times (how I do it every time, to get sure) than I changed the Version again (without canging anything on the scanner itself) and tried the calibration again - and I got a realistic result and also 3 times the same, but it was 0,22° different to the try with 3.8. So what is the right calibration?

Greetings,

Alex

[Sven]

Hi Alex,

The calibration of the NEWER version is probably more precise than the old one. But we still need to test, that's why it's beta. You can assume the calibration is very good when the scans fit well together 360°.

About your problem of "not failing" and unrealistic results: Please send me some images.
Best would be to have your "SL-TestSequence" of that particular case:
1. Create a subfolder "SL-TestSequence" in your DAVID application folder (DAVID must have write access).
2. Calibrate.
3. Send me all files in "SL-TestSequence". It will be large, so please ZIP and use a file hoster or https://www.wetransfer.com/.

Thanks!
Sven