DSM and Ortho resolutions are the same. That being said, when inspecting my vertical accuracy (with my DSM) I get great results (SD=4.8 cm), though my horizontal accuracy (with my orthomosaic) is very poor comparably (30 cm).
Does Pix4D not georeference the images for the orthomosaic very tightly to the detailed DSM? Does it use a lower-res DSM for the orthorectification (which would lead to the horizontal error being high for the orthomosaic)? And if so, is there a way to make it be more tightly controlled?
Bottom line: I’m looking for a way to improve the georeferenced horizontal accuracy of my orthomosaic.
-Addy
Hello Addy,
Thank you for your message.
How do you measure the horizontal accuracy? Do you use the checkpoints?
There should be no difference in absolute accuracy between DSM and orthomosaic.
If possible, could you share a quality report from one of the projects where you observe this issue, or, if you prefer, you can reach out to Pix4D support team for further assistance.
Kind regards,
Alexey
Alexey,
I’m measuring horizontal accuracy in ArcPro using the Pix4Dmatic generated orthomosaic and GCPs. When looking at the quality report I do see they are both 18 cm/px (637x977px) - my question is more so as to why the accuracy of the vertical is better than the horizontal. I’ve attached my latest quality report for you to review.
Project_12_05_24-quality_report.pdf (1.6 MB)
Addy
Hello Andy,
Thank you for your reply and for sharing a quality report. Indeed, the the average GSD is 9 cm/px, so we might expect the average relative accuracy of ~18 cm/px; from this point of view 30 cm/px is higher than expected.
Based on the quality report it is not clear where it comes from, in the quality report there is no GCPs or checkpoints, do you use them in PIX4Dmatic? It might be interesting to add your control points to PIX4Dmatic and to use them as check points, could you give it a try?
Kind regards,
Alexey
Hi Alexey, @alexey.gribovskiy
No GCPs or checkpoints were used - this was intentional as the remote nature of my work often does not allow for on-the-ground GCPs. My goal is to develop a workflow for my data that does not require GCP/checkpoints for accuracy higher than 10-15 cm.
Thanks for your help on this!
Addy
Hello Addy,
Thank you for your reply and for the details.
If possible, could you provide the following details:
- How do you georeference the images, do you use RTK/PPK solutions for this?
- What is the absolute horizontal / vertical accuracy of the images positions?
- What is the ground truth data that is used for the vertical/horizontal verification?
Thank you and kind regards,
Alexey
Alexey,
- We use PPK processed GNSS data from a Trimble R7 with corrections to a local base station.
- Before P4D process our accuracy is as follows: Horizontal - 99.8 % of our PPK derived image positions are within 5 cm, the remainder falling between 5-15 cm.
- We use our own highly accurate GCPs collected with, I believe, Trimble R7s.
I am using GCPs in Arc to assess accuracy, but am not using GCPs within Pix4Dmatic to improve accuracy. This is a test case where we do have GCPs, but in most cases we do not, so our accuracy cannot hinge on GCPs and is simply a validation tool.