CGP's and altitude

[aerial_survey]

HI There,
I am using many GCP's for a large area, but my results are not coming back at a good accuracy.
When I have added the flag to the GCP's like 4-6 I then optimize camera alignment. Should i be able to then check the marker height to the imported CSV height, they are about 150mm different? SDhould the height get adopted to the csv always?
see attachement...
Thanks
Allan

[MaciekK]

Hi,
it looks like all errors have the same sign. So there is a vertical shift of the coordinate system. Probably setting the aprori error for GCP to 0.015m will improve the solution.

Maciek

[Alexey Pasumansky]

Hello Allan,

You can leave some markers in the Reference pane unchecked, thus do not include them to georeferencing and optimization operations, and later check the accuracy of the alignment/optimization results based on the errors observed on those markers (check points).

However, isn't 1.5 cm error considered as good accuracy for the aerial surveys?

[SAV]

Hi aerial_survey,
 
According to your screenshot your markers are at around Z=80m and your DJI drone camera coordinates say Z=185m. The marker errors as listed in the table are less than 2.5cm (largest value is 0.024205m), which is excellent , as already mentioned by Alexey.

For example, the ground sampling distance (GSD) for a survey flown with a DJI Phantom 4 Pro at AGL=100m is 2.7cm per pixel. The rule of thumb is that your error is around 1-3 times your GSD.

Regarding the difference of 10+cm as highlighted in the picture. I think you might have forgotten to recompute the DEM / depth maps after optimizing the camera alignment. Optimization only affects the sparse point cloud, which means it is crucial to run it BEFORE computing  any other datasets such as the DEM, depth maps, dense point cloud or orthophotograph.

In cases where you have already computed the DEM/dense point cloud etc you will have to rerun these steps after doing the optimisation.

All the best.

Regards,
SAV