I have been working on using Emlid Studio to process my drone survey images.
Having managed to get a result and after geotagging the images with Studio, I processed them with Metashape. The PPK seems to have had a strange effect on the height data, giving the resultant model an unreal curved aspect.
You can see two screen shots of the results here - LINK
One image is of the model from geotagging with RTK and the second is the same data, but after geotagging with PPK in Studio.
So far, I have only tried a few datasets with PPK. Some of these were created with RTK available and some without. At least one other model has produced fairly normal looking results from just PPK.
What could be going on here? My understanding is that PPK should produce slightly better results than RTK. I’ve also been given to understand that PPK can be run on an RTK dataset as a check on accuracy.
In this case the RTK seems accurate, but the PPK has introduced a degree of error.
It is called the banana effect in photogrammetry. It mostly occurs with an image dataset where the camera is moved parallel to the object’s surface and where the surface is almost planar or with a very small visible parallax. This error happens due to image distortions and the compensation of internal and external calibration parameters such as camera altitude and camera focal length.
You can read more about it on this website.
The properly geotagged photos should resolve this issue, so it’s curious that you just get a good result with RTK. Could you double-check if you have the geotags in the drone images’ EXIF data after processing it with Emlid Studio?
What drone are you using? DJI RTK drones such as P4RTK, M3E, M300, etc, have extensive XMP metadata embedded in each image, including camera calibration, yaw, pitch and roll angles, and other variables that Metashape uses for optimisation. Unfortunately Emlid Studio doesn’t retain this data when it writes the new images, it just writes the corrected coordinates in the EXIF metadata. This is most likely the reason for your issue.
You can check this by looking at the camera data in Metashape, a normal RTK tagged image straight from the drone will show yes under precalibrated whereas PPK processed from Emlid Studio will show no. I believe retaining the XMP data is on Emlid’s roadmap.
You can try using the ‘fit additional corrections’ option when optimising (although this increases the dense cloud processing time) or ‘adaptive camera model fitting’ in the alignment dialog.
