Laser and Photo Reconstruction

Comments

23 comments

  • Avatar
    Götz Echtenacher

    Hi Spencer,

    never experienced that before and I have done it a couple of times already. This should definitely not happen.

    The orange lines at the cubes that represent your scanner's positions strike me as odd.

    Similar things happen when there is some kind of problem, but I can't recall what exactly. Might have to do with the coordinate system but i'm really not sure. If you search here on the forum you might find the answer or at least give you a hint of what might be causing this...

    1
    Comment actions Permalink
  • Avatar
    Cabral

    I think I've found the problem. It's the minimal distance between two vertex under mesh calculation. It seems that its default value - 0 - works in the majority of the model but not for every area. For these it needs to be changed to a value other than 0.

    I've been testing with 2mm (same as for the laser scanning res) and it seems to work with different sizes and placements of the reconstruction region.

    1
    Comment actions Permalink
  • Avatar
    Spencer Idenouye

    Hi Gotz,

    Thanks for the comments. Just knowing that the issue should definitely not be happening helps. Especially since all the tutorials and documentation just describe the process as straight forward. I'm using Lieca's new BLK360 lidar scanner (which is a relatively new product), so it could entirely be something wrong with what the scanner is doing. There's a new firmware update so I will try that. I'm also doing scan alignment in ***** and then exporting the e57, which could be causing a problem for reality capture. I might try exporting individual scans or have reality capture align them as opposed to importing as exact.

    Thanks!

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Spencer,

    oh, you did not import them as individual scans? Well, then we have the solution! :-)

    The need to be individual...

    I've heard that RCs alignment is not always ideal, but that might be personal opinion.

    Did you **** the name of the app or is it some paranoid screening automatism?

    Test **** [Edit: these stars have been added by RC Support to expunge already partly *ed names of competitors. This shows it's manual and not automatic. For whatever this info is good...  :-)  ]

    No automatic screening - I ****ed it myself after posting it.  :-)

    Interesting...

    0
    Comment actions Permalink
  • Avatar
    Spencer Idenouye

    Oh interesting - there was automatic screening on the other software name. Very strange. Anyway lidar registration was done in **** pro from which I exported an e57. I'll try the individual scans.

    0
    Comment actions Permalink
  • Avatar
    Spencer Idenouye

    I should clarify that I have done 30 lidar scans of a location. Then registered them all together in 3rd party software. Then exported all 30 scans combined as a single e57 "scan". But I guess now I can try exporting each of the 30 scans individually as their own e57 file and try importing those into RC (which seems like a whole lot of extra work).

    0
    Comment actions Permalink
  • Avatar
    chris

    that will be your problem.

    export each location. rc should create 180 files from the 30 scan locations.

    if you do a single file, rc will only make 6 files. and it won't work very well.  

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hey Spencer,

    yes, separate scans should do the trick. And remember to include RGB as well, if you have it.

    About the screening, some people running the forum here are a bit touchy about mentioning direct competitors or really discuss any other software in length. Where I can understand that it's necessary to put a lid on excessive cases, I think it is actually counterproductive in most cases. I know that not everybody at RC agrees with this practice and it's been eased in between, but there you go. The partial *ing is what I've seen the mods do, so it should be ok.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Thank you very much. I edited mine as well.

    In your Code of Conduct it says not to ADVERTISE nor to HIGHLIGHT competitors names.

    If they are simply mentioned in order to describe and therefore solve a specific problem, that is NEITHER. Especially if they are already partily altered to prevent search engines from finding them. In the past, support has left it at changing a couple of letters to stars, which I adopted. Some problems are software specific or at least could be. With partially obscuring the names, people will at least have a chance to guess what is happening.

    If you keep up this over-strict policy, the forum looses quite a bit of it's purpose, which is to a great part helping users with specific problems. The alternative is that, as it has happened in this very thread, people evade to privade communication. I am sure this is not in your interest.

    With Kind Regards

     

    0
    Comment actions Permalink
  • Avatar
    chris

    yeah rc is used as part of a pipeline of lots of other software.

    not sure if its really that helpful for anyone to ignore that.

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Hi Spencer & Götz,

     

    I'm having a similar issue as the one described by Spencer. Also with the same gear (BLK360) and same import settings. So you're saying that it is not possible to align photos and scans datasets in RC if the scans are imported as exact (pre-registered on Recap Pro for instance) ?

     

    Cheers

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Cabral,

    pre-registring and then importing as exact is recommended. That's not the issue.

    RC needs to know where each scan was positioned so you just need to make sure not to combine all scans into one big one but leave them individual. The proper term is ordered pointclouds, I believe.

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Hi Gotz,

     

    Ah I see what you mean now. I believe Recap calls that "Complete" or "3d-only", if I understand correctly. I've tested 3d-only and yes, it didn't work, so I've been exporting the complete data. So I guess that's not the issue affecting my reconstruction problems.

    Thanks for the help!

     

    0
    Comment actions Permalink
  • Avatar
    Spencer Idenouye

    I’ve also been preregistering, exporting an e57 as complete, and importing into RC as complete. Once imported I align (which is very fast since the laser scan is exact) then I can see a nice point cloud. The I set my region and calculate, then import photos and re-align. At this point things are looking good, if photos have aligned correctly they match up with the laser scan. Sometimes they won’t align in which case additional components are automatically created within the project. Another related issue I’m having is that the BLK360 is unable to provide the source photo images that the scanner takes. It often takes suitable photos, then stitches them internally (I think) into a 360 panoramic image. These 360 images can be viewed at each scan location in other software, but when I export as an e57, those higher quality images are lost. When the e57 is imported into RC, some version of images are imported, but they are nowhere near the original photo quality. They look like an image that was created as a colorized point cloud, they are very poor. This makes it impossible to add tie points to realign components from photo sets together with the laser scan. When I continue to calculate reconstruction with the combined laser and photo alignment, the results are very poor. independently the laser scan reconstructs well (just with no suitable textures), and when I reconstruct with just photos that also typically yields good results. But combining the two types of data are not working for me.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Cabral,

    no problem - I hope it works out for you!

     

    Hi Spencer,

    the result with both combined should not be worse than the images alone. If it is, then there is probably something wrong with the setting (or the scan import - see below). The fact that you still get several components confirms that.

    Did you use RGB as I asked before?

    The 360 image wouldn't be of any use since RC cannot (yet?) use those. What you describe is actually correct, RC transfers the pointcloud into 6 images for each scan. Due to the precision of the scan, the resulting depth maps are very accurate. The matching process is from then on the same as from image to image. If the images have too low a resolution for your taste, maybe you reduced the density for export (from your registering software) too much? That could also be the reason for a failing alignment.

    I've never tried using tie points for laserscans. But judging from the screenshot at the beginning there shouldn't be any problems since your images seem to have been taken properly - maybe even a tad close within the circles. In any case they should be suficient to model the plants to a very high detail - much more than you probably need for the model of the whole building.

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Hi Spencer

     

    I'm experiencing the same issues, and it has been driving me crazy for the last couple of days!

    Moreover, I can only get a reconstruction (point cloud/mesh) from the laser component if I change the "minimal distance between two points" to 0.02 (instead of the default 0.002). I imagine that this is due to the medium res scans of the blk360 not being dense enough. Are you able to reconstruct using the default 0.002?

    When combining laser & photos I'm getting this with 0.002:

    and now just tested with 0.02 and got this:

    Did you manage to test exporting each scan location to an individual E57 file? Did it produce better results? 

     

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Just an update:

    I finally managed to get a reconstruction on normal detail by limiting the reconstruction region to a small portion of the entire model. This is the result:

    I will now try to run more tests to see if I can understand why I can't get the entire model to reconstruct.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Cabral,

    this is really strange. I could imagine that you have maybe set a very low maximum poly count for the entire mesh, which would explain the weird results for the whole model but an acceptable outcome for a small part. Although your settings seem to be ok. Just to be sure, have you tried to reset the settings to default? It can be done by shift-doubleklick on start or for each setup menu individually with the three tiny buttons on the bottom left.

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Hi Gotz,

     

    I did a settings reset yet again (I do it regularly when things start behaving strange) and I'still getting the same issues. A few more updates from the latest tests on the component with the aligned scans and photos:

    - I can get a reconstruction if I disable the laser scans (750M tris on normal res)

    - I can get a reconstruction if I disable the photos and set the minimal sampling distance to 2cm (1cm or lower does not work)

    If I try laser + photos with a minimal sampling distance of 2cm I get an Unknown Error

     So this is getting quite frustrating as I'm still not sure if it's a software issue or if it's just a matter of changing a setting. I just wish one of the devs would be bothered enough to give some sort of reply. 

    Edit:

    Ok I now realized that there is also some kind of issue with some areas of the model.

    I reduced the size of the reconstruction region and, as expected, I could get a model with both laser scans (3cm res) and photos:

    but then I slightly changed the region towards the side of the model (without changing the size or the settings) and got this:

    So what could explain this? Too many inputs in specific regions?

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Cabral,

    thank you very much for reporting back! That might help a lot of people, including the TO...

    That definitely seems like a bug to me and I would suggest you open a report.

    0
    Comment actions Permalink
  • Avatar
    Cabral

    Hi Gotz,

     

    Sure, no problem. Thanks for your feedback also. It always helps to track these issues.

    I'll submit a bug report then.

    0
    Comment actions Permalink
  • Avatar
    Spencer Idenouye

    Hey Gotz and Cabral,

    Thanks so much for your advice/trouble shooting!  Changing the minimal distance between two vertex under mesh calculation to 2mm to match the laser scan seems to work for me as well. Going to go back and continue testing some of my previous scans.

     

    Thanks!

    0
    Comment actions Permalink
  • Avatar
    RealityCapture Support

    hello everybody, your posts have been edited according to the code of conduct of this forum: https://support.capturingreality.com/hc/en-us/community/posts/115000790911-Code-Of-Conduct

    -1
    Comment actions Permalink

Please sign in to leave a comment.