translating to local coordinates

[fberna]

hello,

I have flown with the drone -Phantom 4- an area of about 100 X 100 m.

Photoscan works flawlessly in its workflow, aligning all 196 pictures, creating dense point cloud, mesh, DEM and Orthomosaic.

I have take 4 ground points with the total station setting up a local coordinates system. TS was on ENZ 100,200,100.


I have done several times before, but using the Swiss coordinates. Photoscan was placing it correctly.


Now with trying to use the exactly same procedure/worflow with the local coordinates is not working.
As soon as I hit the update button the point cloud gets a very strange form.

The error is very different from each of the four points.

I have tried to run the procedure starting with the GPS coordinates and converting first to Swiss coordinates. In both cases I run into the same problem.

Am I doing something wrong -most probably-? Any hint?

[Alexey Pasumansky]

Hello fberna,

I think that you need to perform the following steps:
- uncheck all cameras in the Reference pane (I assume that they might be in WGS84),
- switch the coordinate system in the Reference pane settings dialog to Local Coordiantes,
- place markers, adjust their projections if necessary, input markers' coordinates to the Source tab of the Reference pane,
- press Update button on the pane.

[fberna]

Thanks Alexey,

your worklow worked well.

For unknown reasons I get a very high error, around 0.5m, which does not make any sense. The ground point were taken with a total station.
I assume that when placing the markers manually on eaech photo we could well have introduced an error of some 3 to 5 cm at the very most.

So I really would like to understand where this large error is coming from...

[Alexey Pasumansky]

Hello fberna,

I can suggest to optimize the camera alignment (there's corresponding button on the Reference pane toolbar) after placing markers and selecting all suggested parameters except K4, P3 and P4. But please note that dense cloud will be discarded by the optimization and you'll have to reprocess also DEM and Orthomosaic generation.

[fberna]

Hello Alexey,

I followed your instructions but I still have a 0.4m East (X axis) error and I really do not understand where it comes from.

I thought I could have made something wrong in measuring the GP, so went back and did it again. Same problem....

Unchecking the marker with the larger error I get down to a mean error of 9 cm, which allows me to measure the stockpile which is very small compared to the all area taken by the drone.

I need to understand where the problem is, because I have other similar project where the error should be kept into the 2 to 3 cm range...

[Alexey Pasumansky]

Hello fberna,

I can suggest to acquire a sample dataset with higher number of ground control points, so that you can have a possibility to use bigger number of markers for optimization and keeping some of them unchecked for accuracy validation purposes.

If you can save the results of alignment/optimization to PSZ project file and send it to support@agisoft.com, we'll try to take a more detailed look.

[fberna]

hello Alexey,

thanks again for your message.

I would be glad to send the data to you.

Do I have to save the project in .psv or just the alignment/optimization result? How would I do it?

[Alexey Pasumansky]

Hello fberna,

Just save the project in PSZ (PhotoScan archive) format after performing the alignment and optimization, no need to build the dense cloud or mesh.

[Alexey Pasumansky]

Hello fberna,

I assume that issue may be related to the rolling shutter effect of the DJI drone camera.
I've tried using the version 1.3 pre-release where it is possible to fix the rolling shutter during optimization and managed to decrease the errors up to a couple of cm average, but with more control and validation points it would be better.

My workflow in the version 1.3 was the following:
- align cameras (using the same matching points) in the chunk with removed markers, adaptive camera model fitting - disabled,
- imported markers, switched the coordinate system to Local Coordinates in the Reference pane settings (as the source coordinate values do not seem to correspond to the selected coordinate system.),
- optimized the alignment with based on markers.

Below are the screenshots of the optimization window and Reference errors after optimization.

But I'm still suggesting to have larger set of control points, as some of them can be used for the accuracy validation.