Added laser Scans, now residuals are seemingly infinite

Comments

26 comments

  • Avatar
    John Joseph

    Here are my alignment settings and reconstruction settings.

  • Avatar
    Götz Echtenacher

    Hi John,

    what happens if you re-align? Or better even create a new project, make sure to import the scan first and then the images. With only 400 images, it should not take too too long...

    I've seen similar screenshots alot lately here in the forum, but can't remember if there was a solution. Might be some problem with the scans themselves, maybe a bug. If you search I'm sure you will find something.

  • Avatar
    Götz Echtenacher

    BTW, are you also posting as Tim B or are there two people working on the same image set?

  • Avatar
    John Joseph

    Tim is on my team :) - just joined 2 weeks ago and came the site when we did the reshoot at. 

  • Avatar
    Götz Echtenacher

    Ah, I see!  :-)

  • Avatar
    chris

    whats your co-ordinate system settings?

    thats probably why residuals are point very far away.

     

    as for the laser scans did you register them before importing into rc?

    it seems to work best if you align them first in something else. and then import into rc as exact. and then align photos to those.

    but I'm interested in seeing the results with the blk360. how has that been to use?

     

  • Avatar
    John Joseph

    Chris,

    I hadn't looked at the coordinate system - where can I find that setting? It's odd because, before the registration step, the residuals looked normal.  

    The laser scans were registered before I exported them to e57 files in another application. I did import them as exact.  I had actually added a set of drone photos first, then added the laser and then finally added some close-up shots.  I'm currently trying again from scratch, starting with the registered e57s, then drone, then closeups.  Currently adding control points to get ortho photos from the drone to connect (needed a bunch last time)

    The BLK360 is interesting. It's pretty easy to use, but currently can't get raw data off it (need to use another piece of software that I'm not crazy about).  I've also noticed weird stitching, which led to doubling of images.  I may need to export individual scans and try to register them in RC as a next step.

    John

  • Avatar
    chris

    for the coordinate system got to workflow tab. then settings.

    personally i use project and output at the same local UTM system. and then i on export I manually set and x,y offset to re-zero the model. 

    this way I get the same export position each time. 

    as for your workflow.

    I would add laserscans first, then drone, then ground level.

    also after you register  scans you should if possible export to many e57 or ptx. so you get one file per location.

    if you export just one file, you lose a lot of info and have had problems with that.

     

  • Avatar
    John Joseph

    Hi there,

    Chris - I'll see if I can figure out how to export multiple e57 or ptx files - rc did end up with 36 las files (I noticed a cube representing the laser scan with each face having its own image

    So as I mentioned, I started again and followed the following process.:

    1. set coordinate system to epsg:2278 / Texas South Central (ftUS) for both Project and output
    2. imported e57 (36 files in one component after alignment)
    3. added drone orthos, aligned, added a bunch of control points to merge 4 components
    4. added handheld camera orbit, added a few more control points
    5. added handheld camera detailed shots, added more control points
    6. got component 0 with 2/553, component 1 with 546/553 (not sure what happened to the other 5 pics, but have seen missing pictures before so assuming it's normal) (see below)
    7. Note no residuals appearing
    8. Normal reconstruction started with an estimated finish time about 50 min ago.  What's interesting is that the progress bar seemed to stop in the same place with the previous attempt.

    I'll let this continue for a few more hours (may just let it chug overnight), but the last time I ended up having to abort and that's when the residuals appeared.

    I think my next step is to try to get individual e57 for each scan (not registered elsewhere) and try to align them in RC.

    If anyone has any advice on this, I'd greatly appreciate it.

  • Avatar
    chris

    how many scans did you do?

    maybe it has worked if you only did 6 scans and it has imported 36 las files.

    it normally makes 6 las files per scan location.

    you can also test the align by selecting it all the cameras and clicking on inspect tool. it will show you the connections between all the cameras. and you look for lots of connections esp. between drone, laserscan and ground photos.

    you can also do this while its processing.

  • Avatar
    John Joseph

    There are only six scans that I brought in to RC.. I technically did 2 more but removed them from the registration due to what appeared to be bad stitching either on the BLK360, or the mobile app (they don't have a scanner specific app yet - forced to used the one by their partner)

    Inspect seems to have no connections between laser cubes and the pictures *see below).  Not what to make of that since they were detected as part of the component...

     

  • Avatar
    chris

    ok looks like it needs more connections, which could be causing issues.

    I suggest not flying the drone manually. you need to make sure you a move a decent distance between photos

    if you can fly a double grid, with camera at 45 degrees. (straight down won't stitch well)

    otherwise multiple overlapping obits or both.

    also for walking around do it 3 times moving closer or further away.

    I would potentially even walk around the laserscan locations a few times each. just so you have some photos close the scans that can then link back to other photos.

    if you can't get back on site. then using control points might help.

     

  • Avatar
    John Joseph

    agree with all of those, though at this point I don't have access to the site again.  I do have other pics from a different day which include some POI autopilot orbits, though the green tape that was added to try to allow the flat white surfaces to be picked up wasn't there the first time out.  I worry that'll be a control point nighmare

    My more immediate concern is why RC is not completing the registration at all.  I was able to get a preview view  (very bloby), but the fact that it sees to be hanging with normal detail is whats troubling

  • Avatar
    chris

    Looks like I lost my last message....

    anyway I would try using all the photos you have on site.

    but you can sort them by exporting image list. (you can also just export one image list with everything and then edit it).

    that way you can select different days to turn on or off, either align, model, or texturing.

    though i would probably try using everything. maybe just adjusting the texturing stage.

    having it work on preview and crash on normal. sounds like its an alignment issue, and the preview version is hiding the issue. 

    but in the end it just may be best to model that part of the scene by hand. it can be re-imported into rc and do texturing on it as well.

  • Avatar
    Götz Echtenacher

    Hi John,

    don't worry about using images from different days. RC can handle differences quite well. I once had a scan from a church where there was entirely different furniture (up to the wall mounted lights) from when I did my shots but it still worked fine. So if there is no change in geometry, there should be no problem at all. RC will just use the taped images for the boxes, but the laserscan should be much better suited for that anyway.

    I also wouldn't worry too terribly much about the missing blue lines since there is a threshold (which you can change). As long as the scans are in the right location and you don't get any noise or doubled surfaces (steps), which is an alignment issue, then you should be fine.

    BTW, I just looked at one of the images Tim uploaded for me and I am afraid that there is way too much post processing going on for any significant detail, it's basically a mosaic. It's better to have more noise than too much processing.

  • Avatar
    Götz Echtenacher

    And I agree with Chris, the camera standpoints are not ideal - you need to think in sweeps (structure from MOTION, as in video) than in panoramas. Everything needs to be connected like a spider web - the more "arms" the better. In addition to circling at different distances, different levels have the advantage of covering geometry at different hights better (e.g. close to the floor).

  • Avatar
    John Joseph

    Looks like it did make it through the normal texturing.

    I'm going to attempt bringing in all the photos (from two separate shoots).  It'll be interesting to see what happens given that the lighting conditions were pretty different (overcast vs bright sun), and the tape isn't on the first set of pics.

    I think I'll also attempt to bring in individual laser scans (ie not registered elsewhere) and see if I can get them to work

  • Avatar
    Götz Echtenacher

    Well, from this distance it doesn't look bad!  :-)

    In my opinion it will only influence the texture, which you will probably not even notice since RC creates an average for each pixel. But you can modify the result by deactivating the undesired images (e.g. the ones with the tape) before texturing...

  • Avatar
    John Joseph

    @Götz  Tim was experimenting with a few different things, hence the different questions on very similar photosets :).  This has helped confirm our own understanding of things beyond the general best practices that we've found online for photogrammetry. Unfortunately, getting to understand all the functions in RC has been a challenge - with the minimal documentation available. Doing our own controlled experiments have been the only way to understand the nuances of RC. This forum has been very helpful so thanks to you and Chris and everyone else that gave us some guidance

    One thing that I was wondering, re. shiny opaque surfaces like metal (say a car) - would shooting with a polarized lens help?  Just another thought for future work.

  • Avatar
    John Joseph

    oh so reconstruct with everything to get the optimal mesh, then turn off the taped images to get a cleaner texture?

  • Avatar
    John Joseph

    I'll likely lock the distance it's viewable from when i add it to the larger scene

     

  • Avatar
    John Joseph

    @Götz  and @Chris - would love to keep in touch offline at some point - john@ondaka.com is my email address - would love to understand what you're both doing with RC/photogrammetry and hear your stories.

     

     

  • Avatar
    Götz Echtenacher

    Hey John,

    I agree with you about the steep learning curve!

    And sorry for confusing you and Tim - I'm sure you understand the potential though!  ;-)

    Every project has it's own needs and what works is always best, but in the case of the "mosaics" I am 99% certain (the missing 1 % is politeness  ;-) that that is the cause for the lack of detail. I positive that with no de-noising and sharpening, there would be enough features on the boxes to get at least the general geometry, maybe with a few flawy. I have good experience with similar metal surfaces (gutters and drainpipes) which are much further away on my shots.

    And yes, throw everything in there for the mesh, but if there are shots too close (as in a few cm), you might get problems from that, so better deactivate one of them.

    I didn't understand your last post...

  • Avatar
    John Joseph

    Just meant that it would be nice to keep in touch outside of this particular forum. I noticed that there are clear rules here about mentioning other photogrammetry products/companies which makes it a risk to discuss pros/cons of various tools or techniques without risking being banned. ie. I have a BLK360, and the only way to get data off of it and to register the scans is with a competing project but not typing a particular 8 character name is tiresome :)

    But no worries either way - I appreciated your help and advice.

  • Avatar
    Götz Echtenacher

    Hi Joseph,

    We were at a time related crossover, I meant the post before tha last!     lol  

    I emailed you already...  ;-)

    I agree with you about the tiresome rules.

    However, I think if the software is no competitor, it shouldn't be a problem if it is in a particular context to a problem related to RC and not excessive. Also, adding 1 or 2 *s usually does the trick.

    To be honest, this thread is one of the most productive in a while so they should be glad for some activity!  >:D

  • Avatar
    chris

    that model is looking better.

    if you mix photos with sun and overcast. you'll get an average of them both. so faded shadows.

    I much prefer using overcast photos, as its easy to re-light in 3d to any time of day.

    but if you have the 2 different days saved out as image lists, then you can try out texturing each set by itself, or all turned on. 

    also i find opening the image lists in excel a good way to sort the lists. it gets over some of the UI issues in rc. at the cost of going to external software which is a bit of a pain. 

     

Please sign in to leave a comment.