This is a puzzlement. I’m losing hair over this issue and at 70 I don’t have much to spare.
I’m doing a comparative processing exercise for a small drone company. Their drone uses RTK GPS with Trimble GPS and attitude equiment. The data is processed through PostPac and yields a SBET and geo.txt file in the appropriate UTM (WGS84) CRS.
This company has adapted a version of ODM 2.8 to accept the geo.txt file ( I think they have gotten their ODM 2.8 based processor to use the omega, phi, kappa values ???) and imagery and gets consistently excellent geolocation results. ( a little less than 2X the GSD size in X/Y error to ground truth). They are using the trianglulation process for SfM. I think this is wonderful for processing without GCP’s.
I’m using the same imagery and geo.txt files from three sites (TN, IL, and TX) and processing through the current WebODM S/W. I am using the incremental SfM. For the TN site, my geolocation error is comparably excellent to the drone company’s processor. On the IL and TX site, the geolocation error is consistently ~1 meter to the east-south-east.