Camera and projector fine adjustments

[Dudemeister]

Taking into account of the correspondence between the projection resolution and camera resolution, in order to avoid the projection of the grid scan data fringe phenomenon, the camera's focus is not together. Also is the manual focus, between the camera focal plane will have a certain deviation, but the depth of field of the camera enough, does not affect to the scan.

This past weekend I did a little bit of work on the scanner, and found that I can easily refocus the projector.

So for this purpose, I first used a camera app to set the focus location for the cameras, and had them focus on a medicine bottle with small writing on it. The left camera was in focus at about 14.5", whereas the right camera was in focus at about 15". Both cameras appear to have <1" depth of field.

I then place a white piece of cardboard at 14.75" distance to the projector and projected a test pattern. The front of the projector lens has 2 tiny holes which can be used to adjust the focus, so I was able to adjust it for a (near) perfect focus.

Once adjusted, I ran a calibration test, and did a few scan. My scans a visibly sharper, although still not what I envisioned. Using manual scanning is better and you can see that individual scans are sharper, but the auto aligning process ultimately softens up the final mesh, probably a byproduct of some fudge factor and guesswork needed to align and generate the watertight mesh.

On a side note I found the INI file which governs the projected patterns and I'd like to know more about it, and how to modify the values. can we get an idea of what's what?

[scobo]

Using manual scanning is better and you can see that individual scans are sharper, but the auto aligning process ultimately softens up the final mesh, probably a byproduct of some fudge factor and guesswork needed to align and generate the watertight mesh.

Don't forget you can save the point cloud data in free scan mode then use other software to create a final mesh.

[Dudemeister]

Don't forget you can save the point cloud data in free scan mode then use other software to create a final mesh.

I've tried that, and I used the Poisson reconstruction filter in Meshlab. Some parts look fine, and in some instances more detailed, but the algorithm used by Einscan is much better at filling in areas with missing data.

So after dedicating almost a whole weekend running various tests and making adjustments, I've pretty much come to the conclusion that the Einscan software is very good at what it does, but it has problems resolving small details, in fact the smaller the scanned object is, the worse the output, but the larger it is, the better the output gets. Probably because it has much more defined (and simply many more) reference points for it's alignment and mesh reconstruction interpolations.

One thing left to explore is the INI file. I don't have it in front of me at the moment, but when I get home tonight, I'll start fooling around with it. From the early look I saw that there are up to 7 steps it can use to project the pattern, but depending on the scan mode or lighting setup, it can use as little as 4 stage. Also there are values describing the pattern (width of each bar, edge blur, etc).

I need to dive into it a bit to see what it does.

maybe you can get more info on the pattern on the David forums ?

I'm not sure the patterns used by the David scanner would be of any use here. David relies on a single camera, and their patterns have both vertical and horizontal lines so the camera can see the deformation in 2 planes. The Einscan uses a pair of cameras producing a stereo image, so it only needs to see the vertical deformation.