Strange huge errors, help!

[Emicimoli]

Hello dear everyone,

I am quite in a rush to hand in a report in the next days and i am having some troubles  hope you can help!

I performed a land based SfM of a valley. When i aligned the pictures everything seemed fine, very nice reconstructio. I located the marker points (which i know very precisely and are calculated using DGPS with 0.001 m accuracy) and Agisoft suggested me exactly the marker spot on other images. However when i add the coordinates, the error is way too high, like about 2 meters with 4 GCPS and 8 meters with all 10 GCPs. This is very weird.
EDIT: Forgot to mention i performed an optimization (at both low values and high), but then the point cloud results in a complete mess !

I used a Nikon D3200 with a 55-300 mm lens. Maybe my mistake was to change the  focal length to often.

I have attached 2 images of the situation.

Do you have any advice on how can i get saved from this situation? I am aiming at an error at least under the meter !

Thanks a lot for your atention

Best regards
Emiliano

[Alexey Pasumansky]

Hello Emicimoli,

Could you please send the project file to support@agisoft.com? Just with the sparse cloud and placed markers. Mesh and dense cloud are not required.

Also some comments on the shooting scenario: taking images from the same camera position is not recommended (unless you are using calibrated nodal head and have grouped images to Camera Station instance in the Workspace pane of PhotoScan window), especially when the distance to the area of interest is much longer then the distance between camera positions. Big focal length may also cause some issues. Suggested focal length interval is 20 - 80 mm in 35mm equivalent.
As for the focal length varying, I can suggest to select all the calibration groups, right-click on selection and use Split Groups option, in this case each image will have individual intrinsic parameters estimated.