Bowl effect / Banana effect and how to prevent it and/or fix it.

[rmckee421]

I am having problems with the bowl effect / banana effect (photos attached). I am using the DJI Mini 4 Pro to do flights at approximately 70-100 m AGL at a consistent elevation. Most of my sites are forested areas following a wildfire. I have attempted to try calibrating the camera prior to aligning photos, which seems to help a little bit. I am not sure if I am doing it right. Agisoft seems to be able to detect the camera that I am using and assign camera calibration values, but I do not know if these are correct or how to find out. I have GCPs collected using a submeter GNSS receiver for some of the flight. Using the GCPs I have been able to mostly eliminate the bowl effect, but not completely.

I am trying to understand what I am doing wrong, and how I can fix it. If I have to purchase a better drone I will, but I'd like to understand what the problem is beofr I spend more money (Agisoft is already pretty expensive and my drone isn't very old). I have read that taking photos at an oblique angle as well as nadir helps, and taking some oblique photos at a different elevation will help. I tried that and it did help a bit, but not very much. I have also heard that it could be a product of the lens size and overall quality of the camera on the DJI Mini 4 Pro. It is also possible I am doing something wrong when I am processing the images in Agisoft, but I find the documentation that came with the software not very helpful to understand or try solutions to problems. My typical workflow to process my imagery includes:
-Add photos
-Calibrate camera
-Align photos (settings in attached screenshot)
-Add GCPs
-Adjust elevation of all the photos (this seems to be a DJI issue not collecting accurate elevations, but could possibly be part of the problem)
-Manually adjust alignment of photos with a sample of GCPs.
-Update Transform
-Manually adjust the rest of the photos with the rest of the GCPs
-Optimize Camera Alignment (I don't really understand these settings so I usually just keep it to defaults).
-Build model
-Build Point Cloud
-Classify Ground Points
-Create Orthomosaic
-Create DEM (using ground points only).

[Paulo]

Hi,

why do you use low accuracy for image alignment?... I would use high with 40000/4000 kpts/tiepts...

If you can share a small dataset, I could have a look at it...

[rmckee421]

Hi Paulo,

Using a higher accuracy alignment often makes the bowl effect worse, so I have been using the lowest accuracy settings possible to limit the bowl effect. The smallest datasets I am currently working with are around 3-5 GB. I'd certainly appreciate your insight though -  I'm pretty new to photogrammetry. How could I share it with you?

[Paulo]

Hi,

you can share with any sharing web (drive, Google drive,....or other. Just the images with any reference info  and calibration data if exists...

[hij]

This can be improved by mixing diagonal photos.
Alternatively, it can be improved by using a camera that has had its lens calibrated.

https://onlinelibrary.wiley.com/doi/full/10.1002/esp.3609

[Braeden]

I use a DJI M3E with RTK GPS.

If I fly with the high precision GPS on, but fail to load the metadata / camera location accuracy then the default 10m location accuracy is used and we get a surface that looks a bit like a pringle.

What camera location accuracy are you using?

[ilia]

Could be something wrong with calibration. What is your calibration settings?