retroactive applicability of RTK imagery

We are considering turning on the RTK capabilities in our eBee now that we know how much work it is to collect, document, and mark GCP’s.  What are our options to use future years’ highly accurate photos to improve the accuracy of our models for previous years that are based on imprecise image geolocation?  

For example, we could use the accurate models based on RTK imagery to identify features to serve as GCP’s using model coordinates, since rocks do not move from year to year.  Would that be appropriate?  How would we estimate the precision of each GCP coordinate based on the model?  This approach still requires a lot of effort to identify and mark lots of GCP’s, but saves the labor of collecting them in the field.  

Are there any approaches that would improve the accuracy of old models based on non-RTK imagery that would not require use of GCP’s?  We probably can’t assume that using the same flight plan would result in identical image coordinates between the two years due to the wind.  Any other ideas?  


If you have time , lets have a quick chat reference your questions.

@Allison, it is possible to take the points reconstructed from the more precise dataset and use it as a reference for previous non-RTK flights.

At the moment, I can think of the following factors that need to be taken into account when estimating the precision of the GCPs extracted from the reconstruction:

  • The precision of image geotags. Typically, these values are determined in real time by senseFly and are written in the EXIF of the images. 
  • A rough estimation of the absolute accuracy of Pix4D outputs is 1-2 GSD horizontally (X,Y coordinates) and 1-3 GSD vertically (Z coordinate) and should be taken into account also for extracted GCPs.
  • The precision of selecting and marking the points in the software.

However, I would always recommend measuring some points in the field in order to have independent checkpoints or backup GCPs if needed and to check if the determined values correspond to the actual values.