We have some troubles to set up correctly the parameter gps-accuracy. We are using RTK data and according to the ODM documentation, the accuracy should be pushed automatically from exif in case of RTK data. We can’t experience this happening so far and we have to set up properly the above mentioned parameter. We have experienced that a higher value like the default value of 10m gives good stitching results but poor geospatial accuracy (~1.5m).
Our GCP benchmarking revealed a geospatiall accuracy of ~10cm and according to Steven, it is recommended to set up 2x the value, so we are using 0.2m gps-accuracy. But in this case, the stitching fails sometimes. My question is:
What is the general recommendation to define the gps-accuracy value, when you get exclusively RTK data from different drones, aiming to ouput a geospatial accuracy of <=10cm for the ortho?
Pls let me also ask, if I correctly understand the parameter:
- At the beginning of the ODM processing you have a process called Structure From Motion (SFM), which calculates the camera positions.
If you have confidence in the accuracy of your GPS positions (=gps-accuracy option), the camera position calculation only allows to calculate camera positions inside of the range, that you set up with gps-accuracy (for example 20cm). If you set up gps accuracy 10m(=default), you give practically SFM all the freedom to calculate the camera positions, which could have a coordinate 10m away from the actual input image coordinate.
I imagine, that GPS-accuracy 10m should work fine, if the camera is correctly calibrated and therefore the output product will have good stitching quality and geospatial precision.
But if it’s not so good calibrated and you have a big difference between the calculates camera positions (SFM) and the input GPS coordinates, a gps-accuracy of 20cm enforces ODM to create an output, that sticks to the initial GPS coordinates in a range of 20cm.
Is this the correct way to approach/understand the parameter?
As always very much appreciated any feedback!