Bowl-shaped distortion in DEM

[Dagon]

I have been making archaeological maps from images acquired with a DJI Phantom 4Pro. Another archaeologist complained to me about the "bowl-shaped" distortions he was seeing in his DEMs. Having worked purely in mountainous sites I had never seen this, but last week I mapped a very flat site, so I finally saw this distortion and it was kind of shocking. I have attached an example.

1) what causes this distortion? I would think no overall distortion could happen because the data comes from hundreds of separate images, so it can't be anything like an anomaly introduced by a single wide-angle lens.

2) Is there an obvious fix that I'm overlooking?

3) Will placing GCPs resolve this? (please tell me there's an easier solution!)

Thanks!

[wizprod]

Have you tried checking the Rolling shutter compensation?
Also, was the alignment optimized?

[Kiesel]

Dagon,

This is a well known effect, caused by the sum of small alignment errors between overlapping images. It is especially big for single line flights with no side overlap.

You can improve your results with
- a good configuration of your image bundle/ mosaic (higher overlapping value, adding oblique and /or overview photos, different flight hight)
- alignment optimisation
- adding GCPs or alternative, accurate image positions by high accuracy GPS.

Regards

[Dagon]

Wizprod: The P4P doesn't use a rolling shutter. I did not optimize the alignment. I will look into that and try it next time.

Keisel: Thank you for the suggestions. I'm currently shooting at 80% overlap, but with the camera angled 90 degrees straight down. I will add some angle and introduce some additional overview photos with higher/oblique shots.

Thanks for your help! I will report back when I make follow-up maps of this site in a couple of weeks.

[wizprod]

Wizprod: The P4P doesn't use a rolling shutter. I did not optimize the alignment. I will look into that and try it next time.

Keisel: Thank you for the suggestions. I'm currently shooting at 80% overlap, but with the camera angled 90 degrees straight down. I will add some angle and introduce some additional overview photos with higher/oblique shots.

Thanks for your help! I will report back when I make follow-up maps of this site in a couple of weeks.

It uses a focal plane shutter, which, depending on your shutter-speed, altitude and speed of the drone, will exhibit rolling shutter. However, it is much better than in cameras without any mechanical shutter at all

[Dagon]

Follow-Up

I flew the same mission again, but this time with the camera angled downward at 75 degrees, rather than 90 degrees straight down. I followed this up with an elliptical orbit shooting at an oblique angle. Processing all of these images together eliminated the distortion I was seeing.

Thanks everyone.

[SAV]

Hi Dagon,

The bowl effect is mostly related to an 'under-constrained' self calibration of the camera. The best way to eliminate this issue is to use accurately surveyed ground control points and a convergent imaging geometry (oblique imagery) as suggested by James et al (2014).

Note that for accurate surveys you should not only rely on geotagged imagery from your P4Pro as reference information.

Even though your new oblique images lead to a reduced bowl effect, you still don't have a good handle on the general accuracy of your model due to your UAV's inaccurate GPS locations, hence I would not use it to extract accurate length/area/volume/orientation measurements.

All the best.

Regards,
SAV

[ArnauCM]

Hi all,

Ironically I am getting the opposite effect. I add pictures with GPS info from the drone, I align them and the model is amazingly flat, almost perfect one would say. Then, I pin my GCPs and add accurate GPS info (0,3m), update and optimize cameras and... Bowl effect.

What am I missing?

[Alexey Pasumansky]

Hello ArnauCM,

Do you include the cameras in the optimization (are cameras checked or unchecked in the Reference pane)?

Also do you see any high errors on GCPs after placing their projections on the images, loading the coordinate information, unchecking the cameras and pressing Update button on the Reference pane (without optimization)?

[ArnauCM]

Hello Alexey,

Yes, I include the cameras in the optimization, as I though that I wanted all of them corrected.

Aparently, there are no errors regarding GCP coordinate info. I can see a small change when checking/unchecking cameras and pressing 'update'; a small tild in the X,Y plane.

[Alexey Pasumansky]

Hello ArnauCM,

Please try to optimize the camera alignment after unchecking all cameras in the Reference pane (while having GCPs checked on). Use all parameters for optimization except K4, P3 and P4, unless some of them are suggested by default (and don't  enable the adaptive camera model fitting option).

[ArnauCM]

As always, thank you very much Alexey!