Hello,
Our group is having trouble with the initial processing step (i.e., we generate uncalibrated images/cameras) when trying to create an orthomosiac using Pix4D. We have generated a lot of UAV-based data on our field using a MicaSense RedEdge-P over the last ~4 years and have been processing it all using Pix4D. One of the issues we have noticed in past years is that some of the orthomosaics generated are blotchy and appear to be missing data. Whenever we go back to the report, we can see a number of red dots (uncalibrated cameras); however, most are blue/green. This year (2024) the number of uncalibrated images was worse than previous years and a stretch of flight dates that we took are rendered useless because of the poor quality orthomosiaic that have been generated. What is strange is that we have not modified the UAV flight path – so its peculiar that some of the processing/calibration work for some of the dates on 2024 (and generate high quality orthomasiacs), but other dates have multiple uncalibrated cameras (the red dots) associated with the initial processing. Visually, when we look at the uncalibrated images using the rayCloud the uncalibrated images “look” exactly like the calibrated images – so to me I don’t think the image quality is the underlying issue. Our group has rerun the initial processing given what is the recommendations of the community here and what we could find online (however, we have had limited success):
- Standard Template
- 96% of images calibrated with a 0.72% camera optimization
- Increase the number of Keypoints from 10,000 > 15,000
- 97% of images calibrated with a 0.68% camera optimization
- Increase the number of Keypoints from 15,000 > 20,000
- 97% of images calibrated with a 1.51% camera optimization
- Adjust the Keypoint Image Scale to ½ instead of Full
- 54% of images calibrated with a 4.73% camera optimization
For comparison a good initial processing typically gives us 99% of images calibrated (3 blocks) with a 0.81% camera optimization.
The real issue is that the Orthomosaics and the Digital Surface Model outputs (that are presented in Figure 1 of the report) look terrible when the image calibration is below 99% (i.e., the 6 orthotactics looks like they have dark lines or scratches across them, while the DSM looks blotchy and the data that is there loses its fine-detail (looks like evelated-blobs); see image below), but looks great when the calibration is at 99% (high quality orthomosaics and crisp DSM). This makes downstream processing impossible with the orthomosiacs that have a < 99% image calibration.
Can anyone give us any support on how we can troubleshoot this, or next steps with the Pix4D software and calibration settings? Thank you!
Best, Luke
