Agisoft Metashape
Agisoft Metashape => General => Topic started by: stihl on March 17, 2014, 01:48:04 PM
-
Hi,
I've finalized generating a 400 hectare land DEM using 67 GCP, all measured with RTK GPS.
Photoscan says the GCP error is 0.036m however when I evaluate the DEM with RTK GPS checkpoints made in the field, far larger errors are found in excess of over 20cm in the Z value.
Is there any explanation as to why? Even exact locations of GCP in the DEM give out false readings when compared to the absolute values of the GPS measurement even though those points should match 1:1?
-
Hello stihl,
Have you optimized the camera alignment based on GCPs only using high marker accuracy parameter in Ground Control pane setting? Also please check that the marker projections have been placed accurately on the corresponding photos.
For quick evaluation of the accuracy I could also recommend to leave some of the markers unchecked in the Ground Control pane, thus they will not be used in the georeferencing and errors on them will show "pure" accuracy for the point.
-
Hi Alexey,
Yes I've optimized the alignment, first by just the GPS camera position (from a Cessna 172) and then after I accurately placed all 67GCP's I've reoptimized it again with all the camera positions unchecked and just the GCP's. In the ground control settings I've set the marker accuracy at 0 since they're measured with RTK GPS.
Photoscan gives an error of 0.036m and 0.768 pixel error, does this sound correct to you?
PS, I'm on the latest version and build.
-
Hello stihl,
Error values seems to be adequate as well as the optimization workflow you've mentioned. However, I recommend to use more real marker accuracy value (like 0.005 or 0.01 depending on the device accuracy), but I don't think "zero" could affect the result in your case.
-
Do you think it's worth starting over from scratch in case something went wrong?
I've unchecked a few GCP's and reoptimized again. Some of the unchecked GCP's stay on the same error(m) except a single pair that went haywire and suddenly gave an error of over half a meter even though the position in the photo hasn't changed. Corrupt project perhaps?
On a second optimization things went completely Fubar.. Unchecked three more GCP's only to see that their respective errors went from a few cm's to nearly a meter, 30cm and 25cm...I've decided to start all over again.
I assume using a camera accuracy of 10m is the best setting for aligning the photo's by just the camera GPS position?
-
I've checked the GCP heights on the DEM that I made. Even though I told Photoscan that the marker accuracy is 0 the height at those exact GCP locations is way off.
Photoscan was talking about an error of 3.6 centimeter however the GCP's that should match 1:1 in height do not match at all, and in some cases the error is up to 40cm!
How is this explained? Photoscan says the error is low, I've told it that the marker accuracy is 0 and thus absolute but it doesn't use those absolute heights for DEM creation.
-
Hello stihl,
Apart from Marker accuracy parameter, Projection accuracy is also used. So by increasing projection accuracy close to zero you'll get errors on marker location also closer to zero. But you should also note that in some cases camera distortion model used could not estimate camera orientations so that all markers fit their positions precisely.
If you wish you can send us the project file with the aligned cameras and markers placed, so that we could try different settings of optimization.
-
Hello Alexey,
It has been brought to my attention that the lens that I used is most likely the culprit of the deformation, thanks to the lens image stabilization that was turned ON.
In the end, GIS works by "Garbage In, Garbage Out". I suspect new photographs will have to be taken.
-
Also be aware that the gps coordinates could have errors. For example incorrect ambiguity resolution if measured using gps.
-
Alexey, how do I "turn off" the projection accuracy, so only GCP's are used?
If I uncheck the boxes next to the cameras in ground control / reference pane, do I need to change anything in the ground control settings pane? Can I leave the camera accuracy set at the default level (10 m)?
-
Hello aggi_user,
If all cameras are unchecked in Ground Control (Reference) pane, then camera accuracy value will not be utilized.
-
Hello Alexey,
It has been brought to my attention that the lens that I used is most likely the culprit of the deformation, thanks to the lens image stabilization that was turned ON.
In the end, GIS works by "Garbage In, Garbage Out". I suspect new photographs will have to be taken.
Yes, this is exactly the point. Your principal point in camera calibration process could not be estimated correctly if you have turned on an image stabilizer. Even the autofocus should be turned off, but has not that terrible effect like the image stabilizer.
You'll have to take the images again.
-
*snip*
Thanks, but this issue was resolved last year. :-)
-
Hi Alexey,
I am getting high errors and I did the camera optimization including GCP's. I dont know if some pictures (attached an image as example) taken upside down are causing these high errors, or maybe I am missing a step in the workflow. I am using pictures from a gopro (fisheye lens) and I already set that in camera calibration but I am not sure if I am doing well.
Can you please shed some light on this matter?
Thanks!
Napo
-
Alexey?
-
Hello Napo,
I see that the camera positions are also used in the georeferencing and have error much higher than accuracy settings, so I recommend to uncheck cameras in the Reference pane.
Also the information about projection errors on markers would be helpful, as well as Adjusted values from camera calibration.
-
Hi Alewey,
Thanks so much for your reply, I will definitely uncheck the camera positions from the reference pane, I thought that using "geotagged" pictures helps with the processing but with that high errors, it probably won't.
I would like to ask you 2 more questions:
I am wondering if the projection errors on markers are the accuracy of the dGPS data?
And, there any standard calibration procedure for GoPro pictures?
Thanks again,
Napo