I am at a loss for how this can happen in processing some drone data.
The project is processing fine for the first section of images, then the data turns down about 90 degrees and the elevation drops, curling the road - where it should be relatively flat. The data has drone and GCP’s I have processed drone data with GCP’s and never had this happen. I have tried multiple options in processing the data. The data in this processing attempt was as follows - Import Photos and GCP’s (Long, Lat, Z). Then I selected photos and GCP matches. I selected unknown for the drone images Absolute Pose option (Previous unsuccessful processing attempts I used the Position for the initial processing, then switched to unknown for the the processing after I assigned the GCP’s to images. The GCP’s at the end are showing a error of 387m. Trying to figure out where the issue is - In the data? (embedded data in the imagery, GCP collection issues…) or do I have something incorrect
in my settings in the software?
Hello, this seems to be a banana effect: Banana Effect - What To Do If My Model is Bent | Tutorial
In the tutorial, there are some advices how to solve such issue.
Were you providing the alignment in the same project as you had after you got this curve?
What you can try is to:
- export control point measurements for GCPs
- open new project
- import images
- set their prior pose as unknown
- import GCPs
- import control points measurements
- align
I have tried multiple different methods for processing this file.
I do have the GCP’s in both Lat Long values and XYZ coordinates and I have processed with both in different projects.
I have also tried removing the GPS data from the drone images and still got the severe banana effect. But that was with just running the initial alignment. I had not selected GCP’s with that data yet.
I am familiar with the banana effect, but as long as the GCP’s are covering the project well I have not seen that occur, it is only when the drone flights extend well past the GCP’s.
I do typically have the initial processing utilize the Position of the photos, then after assigning the GCP’s to photos I will switch to unknown with success in the past, just not with this data set.
The weird thing is the GCP’s getting moved from where they should be located. Also it is looking good for the first part of the data then the banana effect occurs.
The unknow information is the origin of the drone and GCP data. I did not collect this data, it was provided to me from an external source.
Is there a way to lock the GCP positions? I just find it odd that the GCPs that should be in fixed locations are allowing the photos to bend them out of there position and reporting a large error.
Could there be any metadata from the drone that can cause this issue?
I have processed DJI and Autel image sets without any issues, but this set of data is from a Skydio X2 Wide.
Are you processing the data in the same project or are you starting a new project with cleared cache?
Have you tried the proposed workflow with using just GCPs and already measured CP’s measurements?
It can be happening for such long and narrow object, even with GCPs.
You can also try to set the GCP’s precision to zeros (then it should be aligned, but you can lost the precision information about the model).
You can also try to process this datasets in parts (adding first part of images, align, add GCPs, align again, add another part of the images, etc).
I have been starting new projects (but not clearing the cache).
Is there going to be any problems with the CP data that I will need to export from the other projects since that data is showing the banana effect?
Here are some screen shots from processing with the images having the GPS positions removed.
First pass with just aligning before assigning GCPs to images. GCP 100-108 area looks well aligned. From 108-118 is the area that gets the banana effect.
Then I added GCP’s to Images and re processed the alignment.
In how big batches are you adding another images (after 108 GCP)?
According to your screen shot, it looks like GCP 100, 101 and 102 are placed wrongly (as there are quite big errors):
Regarding the exporting CPs, in that case only image measurements are exported, which can be used in a new project for quicker GCPs placing (basically, the images measurements are used for GCP placing, you don’t need to place them manually). And as the GCPs are already measured, you will save some time in a new project.
You can also try to set the GCP’s position accuracy to a lower value (sometimes 0 will help). In such cases also helps to know the exact camera calibration parameters and set them as fixed (which is also not exact workflow, but can help to align such data).
It looks like there is something going on with the data that I was given. I am stopping work with this data. Thanks for the help in trying to sort this issue out, but ultimately you can’t fix bad input data issues.
