I am having a problem with an orthomosaic map I am trying to create. It is of a heavily forested area, and after reading some past posts on this site, I know tree canopies can be problematic. I have tried increasing the minimum point requirement and neighbor matching numbers as advised but am still struggling to get the whole map. Currently, the quality report is is saying its only able to reconstruct 64.5% of the images.
I wanted to see if this was a problem with my images, so I uploaded them to DroneDeploy and got a perfect map. How can I adjust the settings in webODM so I can get a map like the one DroneDeploy provided?
I tried running it again with the default settings. Here is the report. Any suggestions would be greatly appreciated. It’s only reconstructing 60% of my photos which is incredibly frustrating. I don’t think features is the problem as I have raised the minimum to 125k+. Caraway-Ranch-1-21-2023-report (5).pdf - Google Drive
No, this is the WebODM version. You can see a bunch of patches missing and other areas are super blurry when zoomed in. Frontal overlap was 80 and side overlap was 70.
Yes, apologies I should have done that in the first place but not super tech savvy so didn’t know exactly where to find it. Here’s a link to it: console.txt - Google Drive
2023-02-07 22:06:14,393 INFO: 51 partial reconstructions in total.
Even with your 10000 min features, it struggled to get there, with the threshold being reduced to quite low levels to get to the minimum for many images. I suspect that is getting into the image noise, so matches wont be made, or incorrect ones perhaps will, resulting in the large number of partial reconstructions. Raising the min number is not going to help much IMHO.
Can you post a few images on Google Drive? I suspect there is a problem with them, perhaps image blur due to slow shutter/fast flying, wind blowing the trees around, or poor focus?
Here are a few of them in a file on drive. Why would the map work through Drone Deploy if the root of the problem is bad images, though? Mapping Images - Google Drive
The images look quite soft, with branches partially blended into each other, so I suspect soft focus, and maybe some motion blur. How fast were you flying?
Different photogrammetry tools have different requirements and yield different results.
That said, DD cannot perform a complete photogrammetry solution with 70% overlap over a forest any more than OpenDroneMap can. There’s softness to the images, but the underlying issue is sidelap, which softness may compound. DD is likely falling back to a planar approach, much like the field option in OpenDroneMap, but they do cleverly fall back to that approach silently, thus masking issues with the dataset.
Try re-running with the Field preset drop-down set. That will likely give very similar results to DD.
Edit:
I should add: when we talk about issues with a dataset, it is never intended as a critique (most of us have collected a 70% sidelap project over forest), but as an aggregate of the experiences we’ve had here and what works with OpenDroneMap and usually most of the similar, closed source equivalents. We try to combine it with our best guesses as to how to process your data, but also tips for future flights, if we have them.
