objects in Inspire Pro model are vertically slanted

[Radbert]

Hi -

I am having some issues processing a very high resolution (2 mm) survey of a 30m by 30m Arctic vegetation plot that was conducted using the Inspire Pro with the X5 camera and 15 mm lens.  The survey is based on 298 images that were imported with geotags and all of them get aligned in the sparse point cloud.  The forward camera orientation stayed constant during the survey for all flight lines (not sure if this can affect anything).

I notice that vertical objects (vegetation and people) in the dense point cloud are all severely slanted in one direction and the location of the cameras are also offset, while they should be located directly above the point cloud (see attached).  The problem exists both before and after optimizing the camera calibration using five GCPs.  Average GCP reprojection error is also 12.6 cm when I should expect perhaps 1 cm.  I also notice that the flying altitude was estimated at 13.1 m (without GCPs) and 17.6m (with GPCs), while the actual flying height was about 10m. 

I have previously conducted similar surveys with an Sony a6000 on an octocopter and obtained excellent results.  With the X5 camera I notice that the calibration parameters are all over the map when I compare them to two other, larger-area surveys I also conducted this summer (see attached).

Does anyone have any ideas as to what is going wrong here? 

Thanks!

[JMR]

Hello, Radbert: I guess is the rolling shutter effect what what causes this. This drone is great, but the camera is not well suited for photogrammetry.  Flying at such low alt. makes Rolling Sutter extremely problematic. Take into account that for this camera to read the sensor takes 47ms, so if you fly at 6m/s the first row of your photo is taken from a point A and the last row is taken from a point B, 28cm ahead. The only real solution is to stop for taking the photos.
I'm affraid the photos you have taken this time might be useless. Sorry

[Radbert]

Thanks - I was thinking that could be an issue.  We were only flying at about 1 m/s, but this would still be a 5 cm distortion due to the rolling shutter, which is significant in relation to our high resolution.  I wonder what the rolling shutter speed is on the a6000 because we were getting excellent models at 4 m/s.

[SuperNewb]

I have had similar slanted data issues with a standard Inspire 1.  I found it occurred when trying to process small areas.  If you reprocess from the beginning without using the photo gps tags, this can be fixed.  Just make sure the GCPs you  use are correct.

[Radbert]

I have had similar slanted data issues with a standard Inspire 1.  I found it occurred when trying to process small areas.  If you reprocess from the beginning without using the photo gps tags, this can be fixed.  Just make sure the GCPs you  use are correct.

Thanks for this suggestion SuperNewb - I will give it a try tomorrow.
Do you know if there is a simple way to strip off the GPS tags from the photos in Photoscan or specify that Photoscan doesn't use them?

[JMR]

I have had similar slanted data issues with a standard Inspire 1.  I found it occurred when trying to process small areas.  If you reprocess from the beginning without using the photo gps tags, this can be fixed.  Just make sure the GCPs you  use are correct.

Thanks for this suggestion SuperNewb - I will give it a try tomorrow.
Do you know if there is a simple way to strip off the GPS tags from the photos in Photoscan or specify that Photoscan doesn't use them?

To uncheck all tags at once, select all photos and uncheck the first one in reference panel. This will make geotags to be neglected by Pscan. You can clear them as well by right click, but I don't think you need it at all.
This strong slant seems not to be just a matter of RS if you flew at 1m/s, so, what Supernew's suggested may explain it better.
Let us know please.

[mrb]

I wonder what the rolling shutter speed is on the a6000 because we were getting excellent models at 4 m/s.

Are you taking stills from video? The a6000 doesn't have rolling shutter for still images, it's a mechanical shutter.

[Alexey Pasumansky]

Hello Radbert,

The forward camera orientation stayed constant during the survey for all flight lines (not sure if this can affect anything).

For the similar projects when the described flight mission was used, we have observed the similar issue like you are describing.

I can suggest to align photos with the fixed calibration option (keeping default calibration parameters) and then switching off the fix calibration flag and optimizing the alignment (maybe cx and cy should be kept unchanged in this case).

[Radbert]

OK, I think I figured this out - it is a rolling shutter issue that was made much worse by our survey strategy of keeping the UAV/camera orientation constant during the entire survey.

Since the camera reads lines from top to bottom, if it is always facing the same forward direction during flight all photos would be similarly compressed due to rolling shutter so this distortion should be at least partially removed through internal camera calibration.  But if you have the UAV travelling forward and then backwards relative to the camera orientation, photos in one flight line will be compressed and in the next will be stretched. 

The solution then is to process two camera calibration groups separately with each containing alternating flight lines.  I just tried this and it solved the problem.  All cameras are now positioned directly above the surveyed area, vertical objects appear vertical, and GCP reprojection error is lowered to 6 cm (or 3 cm if I implement the new rolling shutter correction in the v1.3 pre-release).

Thanks all for your ideas!