AT Failure during Bundle Adjustment

Out of curiosity .. does it AT without the ground control? My usual workflow - as I don't have RTK on the drone - is to run the AT first - then add say four control points in enough photos to get it georeferenced .. then add more/the rest.

Also did you do a calibrate run with a few photos? So settings for the big AT then would be "Keep camera calibration" .. I usually change to high tie points ..

We use manually installed GCP's since we do not have RTK on our drones either. I ran the AT with the GCP's due to the huge dataset which takes a good few hours to run AT on our computer. Currently processing the AT with the "Keep camera calibration' settings and will see how it goes.

A trial run of a part of the model without GCP's ran fine (AT+ Production). Our dataset mainly consists of grid patterns of 90 deg and 45 deg taken with a DJI GO4 app and manual orbits around some of the structures. I wonder if combining the manual orbits with the grid patterns is somehow affecting the overlap? I am just spit balling.

Single drone or multiple?

From where you are at - I'd do the following:

1) check out the post and confirm your imagery complies with everything at camera calibration

2) get a subset of your images over a "complex" area - like a building - so all four obliques .. nadir.

3) create a new block - call it "Calibration" will do

4) add photos you have selected from (2)

5) run an AT on the "Calibration" block per the calibration post (1) - so multi-pass on optical calcs - high tie points

6) once that is run - export the optical properties to a file - keep them with your job

7) clone your big existing block - set "additional" parameters to "structured aerial data set"

8) import optical properties on to the new clone

9) run an AT - "Keep camera calibration" .. and set tie points to high

10) how did it go?

If you have multiple drones - this needs to be done for each unit and the photos need to be split - each group will have slightly different focal lengths etc - even if the same model of drone.

Did you have auto-focus or manual infinity focus? Need infinity - using AF changes the focal length per photos (perhaps)

Typically when I have done something like this - I'd do each flight as a separate block - so "Nadir"; "North Oblique" etc .. with the survey control and so on ..

Then I merge the blocks and (can) merge the photos groups per drone - and run a final AT.

Having said that - sometimes I have _FELT_ (no empirical evidence) - that the separate photo blocks per "mission" (eg per oblique) has resulted in a better mesh .. rather than trying to fudge the final AT - you might like to try a few tiles per and post final merged AT and decide for yourself.

Single drone or multiple?

From where you are at - I'd do the following:

1) check out the post and confirm your imagery complies with everything at camera calibration

2) get a subset of your images over a "complex" area - like a building - so all four obliques .. nadir.

3) create a new block - call it "Calibration" will do

4) add photos you have selected from (2)

5) run an AT on the "Calibration" block per the calibration post (1) - so multi-pass on optical calcs - high tie points

6) once that is run - export the optical properties to a file - keep them with your job

7) clone your big existing block - set "additional" parameters to "structured aerial data set"

8) import optical properties on to the new clone

9) run an AT - "Keep camera calibration" .. and set tie points to high

10) how did it go?

If you have multiple drones - this needs to be done for each unit and the photos need to be split - each group will have slightly different focal lengths etc - even if the same model of drone.

Did you have auto-focus or manual infinity focus? Need infinity - using AF changes the focal length per photos (perhaps)

Typically when I have done something like this - I'd do each flight as a separate block - so "Nadir"; "North Oblique" etc .. with the survey control and so on ..

Then I merge the blocks and (can) merge the photos groups per drone - and run a final AT.

Having said that - sometimes I have _FELT_ (no empirical evidence) - that the separate photo blocks per "mission" (eg per oblique) has resulted in a better mesh .. rather than trying to fudge the final AT - you might like to try a few tiles per and post final merged AT and decide for yourself.