Wrong camera orientation after photo alignment

[chanmaoc]

Dear Metashape developers and users,

I am working on a project on the rainforest canopy in Peru. We used a DJI L1 drone to capture both LiDAR points and optical images always looking vertically down.

During the flights, we had a ground RTK station, and I used it to do PPK correction using RedToolBox. The EXIF of each image was rewritten during the process so the Metashape can read the corrected camera information directly (I checked the option in the advanced setting).

After photo alignment, I found some of the images (135 out of 1295) could not be registered (Fig. 1). To figure out the issue, I checked the unaligned images alone by deleting the aligned images and redoing the photo-alignment process. What I saw is that their camera orientation is surprisingly wrong (Fig. 2), even though their metadata looks no problem as the angle of pitch and roll is 0 or near 0 (indicating looking to the ground) (Fig. 3).

Does someone know what happened and how to solve this? Thanks!
 

 

[mrv2020]

Hi, I could be wrong, but the capture process has low coverage for the photogrammetric alignment process.

I frequently use DJI L1 and recently L2 in areas with dense vegetation. A big mistake that happens in the process is the inadequacy of the relief to maintain ground parallelism for the flight, excessive speeds and especially winds in the treetops. at the time of capture.

Separating the processes, Lidar captures perfectly, but for photogrammetry it leaves something to be desired.

If you are unable to fly again, try using markers to help the alignment process and also consider aligning at a lower accuracy than usual to force the alignment.

I hope it helps,

[chanmaoc]

Thanks for your reply! It is quite helpful to know where could go wrong and how to improve in future flights!

Agree - the issue may come from the low coverage between photos. Our photo overlap ratio was set at 70%, a ratio that works for flat terrain but is likely too low for dense vegetation. We just realized that. A study (https://www.mdpi.com/1999-4907/12/2/250) used 90% overlap for their small-scale tree size survey, and it looks performing well. Next time we may also try this ratio.

If we can't get a chance to revisit the site, I will try to manually add markers, though I guess you know, that's quite time-consuming!