Computed camera position wrong, project tilted

Dear Community,
I am working with a large amount of images (>50 000) which I would like to process to one single mosaic. The images are taken with an aircraft with one NIR camera and two RGB camera (one rightlooking and one leftlooking). I have split up the images into subsets (with circa 2000 images in each). When I run the initial processing the NIR cloud is not aligned to the RGB project and the computed camera positions differ a lot from the initial (screenshot 1). I run the project on a smaller subset (500 images) and they align perfectly and the computed and initial camera parameters match. The only parameter I changed for the small subset was the Keypoints Image Scale (1 instead of 1/2) and I grouped the RGB cameras into two groups (left and right).
The red marked area in the first screenshot represent the small area processed in the second screenshot. I have attached the quality report of the bigger area.
Any ideas what could possibly be wrong?
Thankful for any help!

region1_5th_Subset_JH_1_report.pdf (1.8 MB)

For this type of homogeneous area, we would recommend the ag RGB template. That will help keeping the image positions correct. However, since you have 3 cameras, we would recommend processing them as separate projects.

1 Like

Hello, thank you for the tip with the RGB template, I will try that out. The reason why I didnt process the different cameras into separate projects is that the camera model is the same for the RGB camera, its just that one s looking to the left and one to the right. So I thought if I do them in separate projects I will not have so many matches and I will also produce mosaic in stripes. I also read somewhere that it is good to process them into one project since then more keypoints can be found which helps to build the 3D model.

I have processed a subset again now, this time with more than 4000 images. The only parameter I changed was the calibration method ( I changed it do geolocation based). The result is much better (see first image). But in the preview of the mosaic it looks like a chess board (image two) in the upper left part of the area, what can be the reason for this?

Is it present in the original images? The stripes in maps mostly result due to sun directionality or low overlap.

It happens when the drone faces the sun in one direction and turn its back to it on the way back. In such a case the angle to the sun changes.

The suggestion would be to fly perpendicular to sun direction and not too early or too late, to avoid the sun from a low angle (around noon). Increasing overlap and flight height would also fix it to some degree.

Can we get your new quality report?

Hello,
yes some images are darker than the other so probably thats why I get this effect, would you recommend me to exclude the flights that are diagonally over the corridor flight tracks?
Here is the quality report:

region1_3rd_report.pdf (1.4 MB)

In the preview of the DSM the result seem to show some stripes, will this go away after further processing?

Kind regards

Hi, You are right, if you have some darker and light images, that will definitely cause this. Pix4D does color balancing for the orthomosaic, but if the exposure is too different, you will see the stripes. I think you already have a very good overlap, so it would be okay to remove the diagonal images. We recommend a uniform overlap.