Im not sure what Im asking is even possible, only just supported the community through the webODM option and attempted something perhaps not straightforward or normal so fully appreciate help may be difficult to come by.
I have a ROV running lines like a drone at 5m altitude with still images of the seafloor and maybe 60% overlap alongtrack. Ideally I was looking to mosaic to create a larger tiff but have been amazed with the photgaramtery outputs which are great and got more than I expected when conditions are good. When going over some shipwrecks etc and in very clear water the results are so much better than I expected, the 3d reconstruction actually very accurate. Sadly the bulk of the seafloor is nondescript so obviously struggles to stitch.
The thing is I have very good positioning (NO EXIF data), not RTK but close over so I wrote a quick script to create a “geo.txt” file and after some teething problems this is now read and I hope used, certainly I cant see errors and tiffs now appear in the correct place. Unfortunately the bulk of the vehicle track does no get generated.
I have not included Omega, Phi and Kappa values although could if I believe this would benefit me by essentially laying the images done without a match. May take e a minute to work out how to calculate them so wonder if its got a chance of working,
If I include these and when working in areas with bad visibility or no targets (a lot of the seafloor looks identical) will the software trust the geo.txt file enough to just lay the files on top of each other without matches?
As the conditions are bad I dont expect the system to be capable of mosaicking simply georeferencing, or is this something thats not possible?
Ive got one folder here called engine. Its an engine on the seafloor. Ive generated a geo.txt file but only populated with lat/long and depth. I have heading (to north) and roll and pitch values for the vehicle on each image but not sure how to transpose to omega phi kappa. I guess heading should be very close as she flys very close to 0 for roll and pitch but otherwise Im not sure how to calculate the correct numbers and didnt seem to do anything.
This engine example you see will come out really nice but I will upload one of a wreck which works on the wreck but the rest of the line is gone.
Thanks Craig
Hi again. So yeah thats me uploading the small shipwreck now also. Dont get super excited its just an old little craft with a tiny engine but still interesting.
I ran the dataset with default settings, it seems to works just fine. I have some concerns on the accuracy of the model, but for general purposes is great.
Photos can be enhanced by adjusting contrast and brigthness, but the best is to improve the mission planning by taking oblique shots of the underwater structures.
Hi Isreal,
Thanks for taking the time. Yes the wreck comes out rather good. The problem I wondered about was how to just lay images flat when nothing can be stitched just trust the geo file. I need a 2d map more than anything.
The dat I sent you will see was a long straight line but it only generates around the wreck,
Thanks again and glad you liked seeing it
I am running the data set without GPS log file ending up with about 15 photos of only 1 MB each which are 5 m above the surface. The 2D shows very clearly the shadow of the white bent pipe(about 90 bend) in the center photo.
The 3D model clearly shows the distance between the pipe and the bottom of the boat estimate by eye say a foot or two. Sides of the hull are rendered nicely, not fully, see vertical clearance from the sea floor even though taken only nadir. Small items on floor render very clearly color is poor only 8 bit but shows bright and good clear/crisp shadows…Hats off to the Team Developers !!!
Problems: 3D extreme bowing …expected
point cloud looks like 3 different boats rotated at different angles…apply texture good solid one boat
Render Time about 3 hours
AMD Threadripper using 48 cores 52 Gig memory
