How is the EXIF data from Drone/UAV/UAS obtained by the on-board GPS

[fiachradunne]

I have a Phantom 3 4k UAV and I use the EXIF data to align my photography in Agisoft. Does anyone know how exactly the geotag is established for each photo? Does it take a new coordinate each time a photo is taken and if so how does it know where its position is if it is moving and tilting as it does not have differential GPS?

[JMR]

The short answer is it not exact at all. The accuracy you might expect from the Phantom's built-in gps (geotag) is about 2m. So the resulting models, if nothing else enters the equation, sould never be considered as accurate maps.
You must use Ground Control Points full stop. But even if you do so, you should take care of a generally ignored issue that is caused by its rolling shutter technology. It partially ruins the internal metrics of an image taken from a fast moving platform and can yield to unacceptable results.
Any serious photographer should be horrified by phantoms and by the croud using them without knowledge about mapping, surveying or photogrammetry.

[jwoods]

The short answer is it not exact at all. The accuracy you might expect from the Phantom's built-in gps (geotag) is about 2m. So the resulting models, if nothing else enters the equation, sould never be considered as accurate maps.
You must use Ground Control Points full stop. But even if you do so, you should take care of a generally ignored issue that is caused by its rolling shutter technology. It partially ruins the internal metrics of an image taken from a fast moving platform and can yield to unacceptable results.
Any serious photographer should be horrified by phantoms and by the croud using them without knowledge about mapping, surveying or photogrammetry.

I've used the Phantom 3 extensively for small scale mapping and the rolling shutter is a non-issue.  At the altitude at which it's useful, between 100-200 ft agl,  it simply doesn't go fast enough to create a shifting effect as long as you have decent light.  My shots usually expose between 1/500 and 1/1k of second.  You would have to fly below 100 feet at it's full speed in low light to see negative effects.

[John_4dm]

There is a solution to geotag photos with 3cm accuracy for almost any drone with no ground control, see this post: http://www.agisoft.com/forum/index.php?topic=5884.0

[JMR]

Hi jwoods. I'm sorry but I have to say you are basically wrong.

The rolling shutter effect, actually has nothing to do with exposure, If you fly at 100ft at full speed you can expose at 1/16000 but your photos will be severely distorted (not shifted). The sensor readout of your Phantom takes as long as 33miliseconds regardless the exposure speed. (I'm not inventing this number, it comes from the specs stated by the sensor manufacturer)
Now lets make some numbers and then you may ask yourself if all this small scale mapping has been as good as you used to believe.
A Phantom3 photo file has 3000 rows
max speed is 16 m/s
if first row is taken when "trigger" starts and last one 33ms later, then your P3 has moved 52,8 centimeters! yes, it is that much!

a perfect square in the ground captured in your photos, will look like a rectangle 52,8cm shorter or longer depending on the sensor readout direction.

But all this means not only that the ground objects will render distorted in your photos, but also means that everything else gets way more complicated. What the camera center coordinates are after all? where the ray intersections are located? common photogrammetry principles are broken because the point of view does not exist as a point.

So, if you want to keep doing extensive mapping with the P3 its fine, but you should wonder if your results are good enough for being considered as professional. One thing is sure, they are nor as accurate as they would be if you made use of a camera with a global or at least mechanical shutter.

Best

[cadm8]

Hello, there is a published paper https://pix4d.com/wp-content/uploads/2016/05/pix4d-isprs-paper-rolling-shutter-final-edited.pdf that actually states that the RS effect from an Inspire with a FC300X camera, ie close to that of the Phantom 3, only with slightly narrower FOV, is about 3-5cm greater when their RS compensation algorithm is not being used. What are your thoughts, could you minimize the effect by going slower at about 3-5m/s?