Getting good scans (Page 2 of 3)

pirvan,

The Raspberry Pi does not have enough memory to efficiently perform high resolution surface reconstruction onboard with an algorithm such as Poisson (however this possibility is much more feasible with the Raspberry Pi 2).  As a result, FreeLSS performs very basic meshing and allows users to create water tight models with 3rd party applications such as MeshLab.  Meshlab also enables noise removal (model smoothing), and other useful capabilities.  The size of the Poisson reconstructed file that MeshLab outputs will largely be driven by the Octree Depth and Solver Divide parameters of the Surface Reconstruction: Poisson dialog.  Increasing these values results in a larger and higher detail model.  A depth of 8 and a divide of 7 are good ones to start with.  Be advised that setting them too high can cause Meshlab to run out of memory and crash.

ugh... do everyone a favor and use something like zippyshare.com  to share files, not whatever that is... it spawned a bunch of popups, and that's with an adblocker on.
oh, and if you are looking to smooth out those "wrinkles", try autodesk's Meshmixer. the "sculpt tool" is especially helpful.

My adblocker kills the zippyshare popup before it happens, but the other site has a script that runs to force a popup, and if you block the script, it doesn't download.

oh, and you really should try meshmixer. I obviously don't have the original model to copy, but here's what a few passes with some of the sculpting tools gave me - I probably smoothed a bit too much, but it's really just an example.

I'd suggest making scans of some really simple objects - a cube, a cylinder, etc. - in order to really see what's going on with your scanner. Having complex shapes, colors, different surfaces and concave areas makes it harder to see what's up.

That said, my scanner is currently sitting idle because I have never been able to get past the problem of the two shells, one from each laser. The latest software update makes the scanner so slow as to be unusable, so I'm hoping that a future release will fix both the speed and the dual-laser scan issues.

UPDATE:  Scanning discrepancy fixed... hopefully.

Yesterday, I went back to scanning simple stuff like the aforementioned medicine bottle, in another attempt to see why the discrepancy between lasers exists, however, midway through the scan I was looking at the preview, and I noticed something I  hadn't noticed before.  The top of the left laser scan is outside, while the bottom is on the inside.

That told me immediately, the left laser is not vertically aligned, yet when using the calibration thingy, it looked fine.  So I printed a checkerboard pattern on a some heavy photopaper stock, and made myself a taller calibration block. 

When I put that on the turntable, i found that both lasers were off, and they were drifting farther apart as they got higher.  Anyway, to make a long story short, once it calibrated both lasers, and the camera to be square, the scanner discrepancy went away.

Here is a pretty complicated model scanned in default mode.  Next up will be the Hi-res scan.

FYI.  The reason I hadn't noticed tilt before, was because FreeLSS, cuts off the overlap between the 2 lasers when the scan is completed.  You can see the offset, but it's impossible to see the cross-over created y the tilt, so you assume the outside at the top is from the same laser that crated the outside at the bottom.

Here is an updated calibration, done at 5Mp.

I also made a small change to my calibration device.  The paper, by itslef, was too flimsy, so I glued the grid to the face of a CD case.  Now it can stand nice and square all by itself.

The fist picture is of the camera by itself, but it's a screen capture, because saving the actual image doesn't include the cross-hair overlay.  The next pictures, in order, are the left laser only, the right laser only, both lasers and the test shot of both lasers.

On another note. 

I tried to do a high-res scan of the little statuette of the lady with the basketful of cats (see image in previous post), and it failed during the merging of the scans from the 2 lasers.  It actually locked up and the scanner was no longer responsive, so I had to reset it.  Needless to say, I lost all the data, the only option I had was to delete.  No view, STL or PLY options, just a delete button.  All that after a 220 minute scan.

I guess scanning something really complex at 3200 steps generates way to much data for for the Pi to process.

That's my thinking too, so right now, I'm doing just that, a 1600 step scan.  As soon as it's done, I'll post the result.  I saw a post somewhere that you were just about done with a new update that might fix the long scans, or was it some sluggishness in the UI?

I never had any issues with sluggishness, but the scanning speed could use some improvement.  At the moment, the scan I'm doing will take about 105 minutes to complete.

BTW, how does the calibration look to you?

I think I finally got this thing dialed in.  Here is the Cat Lady statuette I was trying to scan.  This was done at 5Mp Still & 1600 steps..  It took 103 minutes to complete.  Now If I can only figure out how to scan the top so I can get the missing parts, I'd have it made.

On to the pics.  The first image is the point cloud generated by the scanner, the other three are the mesh generated by the Poisson Surface reconstruction filter.  I then ran a Quadratic Edge Collapse to simplyfy the mesh, and reduce it to about 1/10th of it's original size.

Besides that I just cut the bottom off, and deleted the ring (the edge of the turntable) around the model.

Obviously it still needs some work, like fixing the hat and the basket handle, but considering the complexity of the model, it's not too bad.

I've been really trying various settings and options while scanning and you can see some of the results on my other thread.

However I'm still struggling with various issues, and one that I came across is that that FreeLSS (or the Raspberry Pi) either run out of resources or memory, as you keep using it, resulting in very strange scans with missing "layers".

When I say "layer" I refer to each snapshot that the scanner takes for every rotational step.

I noticed that after a few scans that look fine, the scanner begins to miss or drop every few "layers", resulting in something that looks like this:

The only solution I found so far was to turn the scanner off, then back on.  If this is indeed a problem with memory and/or resources, perhaps Uriah can provide a memory clear button or maybe a simple reboot option.

I think I may have found a workaround this laser discrepancy.

I started playing with the only setting that is not "interconnected" with the others, the "Laser X".  By default the Laser X is set to -4.95" (left) and + 4.95" (right).  My actual dimension from the center of the camera to the center of the laser face is 5.125".  The face of the laser barrel is flush with the plastic holder.

Anyway, I used the actual dimensions, I tried default dimensions, but in either case, the result ws the same, the left and right lasers see a different picture, so this time I input a different number in the left laser, sine it appears to be the one that is off.  I first used a smaller number -4.5", but it didn't appear to make any difference, so I put in -4".  The distortion got worse, so I went the other way.  After some trial and error, I found the numbers to make the lasers agree (almost perfectly):  + 4.95" for the right laser, and -5.3" for the left laser.

I don't know why this works, the numbers do not match actual measurements, but it works.

I'm doing a few more scans at the moment, and I'll post same samples later

Well, it looks like the laser misalignment has been solved by Uriah.  The newly minted FreeLSS v.1.7 solves the problem with it's auto-calibration feature.

If you had similar problems as I did, check out the new version.

Be aware that there is a "glitch" with the camera preview being offset to the left.  Do not realign your camera, leave it as is.  As long as your lasers point to the center of the calibration tool, you're good.  Ignore the camera view red cross-hairs. 

Click the "Calibrate" button, wait for FreeLSS to give you a successful calibration message and start scanning.

Earlier versions of the software would try a particular frame several times if it thought that something was moving in the background (it would wait until the disturbance went away).  This led to a lot of false detections and slowed the scanning process for a lot of people.  Thus, it has been disabled in the most recent code base until it can be revisited.