Tilted Model

[vtrugby]

Hello,

I am using three flights with around 100 photos per flight.  I have geotagged all of the photos before adding them into agisoft.  Is there any common problem that makes models turn out tilted.  I make contours from the model and the elevation of a roof will often be the same as the ground around it.  Tilted at 70 degree angle. 

Any help with this would be greatly appreciated.  Thank you,

[vtrugby]

For more information.  I am using a DJI Phantom with a gorpo flytrex gps logger.

[Alexey Pasumansky]

Hello vtrugby,

Can you post a screenshot of the model view with the camera positions (so that coordinate system axis are visible)?

[vtrugby]

Attached is a photo with axis to show the tilt.

[Alexey Pasumansky]

Hello vtrugby,

And what about the camera positions? Is it a linear flight? Maybe you can also send the report file from this project to our support?

[vtrugby]

The flight is linear.  It is done using DJI ground station photogrammetry tool.  Grid pattern. 

When I look at the camera positions they match how tilted the model is.  Considering during the flight the UAV stays at a constant elevation I am not sure how the camera positions are so tilted.  I would assume the gps data for the DJI Phantom is correct, or else it would be flying weird pattern.  What is the report file?

[Alexey Pasumansky]

Hello vtrugby,

You can export report using File Menu -> Generate Report option.

[vtrugby]

Attached is the report

[vtrugby]

When I load the pictures in the first row on the grid pattern in much lower in height than the others.  I checked my gps data and it varies from 490 feet to 650 feet.  When I fly the telemetrics tell me I am at a constant 49-50 meters height.  I can't imagine the gps system is off since it is flying at constant height, but I don't see how the logger would be logging the gps data in error and creating different numbers.....  Either DJI gps isn't accurate enough (which I doubt) or the logger is going crazy.

P.S. I have previously created great models that had contours matching work done for thousands of dollars.  Also I made sure to calibrate the compass.

[Alexey Pasumansky]

Hello vtrugby,

I can suggest to switch camera type from Frame to Fisheye before aligning images, in case the images are taken with the ultra-wide angle lens.

As for the altitude values, DJI may be saving in EXIF above ground level, while PhotoScan expects height above ellipsoid. You can check the image files themselves, in the header they can store Reference Altitude and Absolute altitude values, but EXIF contains only one of the values.

[vtrugby]

Alexey,

I appreciate your help with this.  After working through it I believe the GPS Logger is at fault or the transfer from DJI GPS to the Logger.  When it tells me I landed 160 feet higher than where I took off, and they are both the same location ...... 

I think since the first row of camera position says it is low, Agisoft tilts the positions to correct it so they are in a plane.  This create the tilt which is no fault of Agisoft, but you helped me learn something with the ellipsoid statement.

Thank you for all the help,

[Alexey Pasumansky]

Hello vtrugby,

If you are not confident is the positions of some cameras you can either uncheck them in the Reference pane, or set lower accuracy value for those cameras (right-click on selection in the Reference pane and choose Set Accuracy option from the context menu). Then press Update button.

[vtrugby]

This seemed to work out really well.  I will make contours and see if they are correct and get back to you on if this was a fix for what is truly wrong.

[vtrugby]

So I did as you recommended and unchecked some of the cameras that positions were off.  This worked great for each flight done as an individual chuck (the contours seems correct). 

I have tried aligning and merging the chunks but all I get is the tilted model like before.  Also doing all the flights at the same time with some of the cameras checked off still produces the tilted model.

I am confused as to why all three models look correct when done individually but when I try to merge them they tilt.  I will try a combination of 2 to see if one of the flights is throwing the others off.

Thanks,

[bigben]

As Alexey pointed out. Gopro = fisheye lens.
I've done some reconstructions for people with similar drones and find that it's often easier to not use the built in GPS metadata. Differences between flights a few hours apart (especially altitude) were problematic