Which drones include corner coordinates in image files?

[BassFromAbove]

Hello all,

The DJI Mavic Air 2, or at least the one that I have which is about three years old now, does not include corner coordinate information in the EXIF part of the image files that it creates. I've seen corner coordinates in examples of other image files from other drone makes and models. I've tried calculating corner coordinates from GPS location, altitude and camera properties with little luck. I've checked the DJI support site and docs but there is no mention of corner coordinates anywhere that I can find. Does anyone know if there is a way to get this drone to include corner coordinates, or have any recommendation of a similarly priced drone that does include corner coordinates?

Thanks much

 

[Yorkshire_Pud]

Can I ask why this is such a necessity?
What gimbal tilt are you considering?

The following is based on thinking about a near horizontal gimbal and would probably be less important if you are talking about shots taken with the gimbal pointing vertically downwards but.......
To my mind it would be a rather large task, as you have seemingly already discovered.
I obviously don't know what you have attempted but to me it appears that you would be basing your estimations on drone AGL, gimbal tilt, roll and yaw/bearing, the camera's field of view and the optical distances to the corners. I.e seemingly some fairly involved 3D trig. Easily performed, I suppose, by a computer program but still involved.

Even with accurate GPS for the drone, your height is relative to the home point and it is not AGL .... unless you have a GPS to ground elevation database etc..
Extending that idea the optical distance to any corner is dependant on not only the drone's AGL etc. but also the corner's 'height'.

Are you sure the examples you have seen give accurate corner co ordinates? It seems to me there are an awful lot of sources of errors.

If you are talking about a few shots rather than mapping an area, have you thought of locating the corners via comparison with the likes of google's etc. satellite views.

 

[BassFromAbove]

Thanks Yorkshire_Pud. The goal is to seamlessly and easily use top-down imagery with zero degree gimbal angle and an AGL of about 50 feet in a custom GIS web application. The application covers land areas ranging from a hundred square feet to multiple acres, so the number of images involved will be one to very many.

For the map bit of the web application, using web/slippy map technology du jour, aerial photography is chopped into tiles for various zoom levels and then integrated with base-map tiles from a common source (e.g. google maps.) I have the tiling process and map integration working fine, it's just getting the coordinates exactly right in order to have the custom map tiles line-up perfectly with the base-map imagery. I can get somewhat close by manually setting reference points on the drone imagery (QGIS et. al. and GDAL) as you suggested, but it is time consuming and rarely able to produce a perfect result. I suspect that the desired AGL is low enough such that the potential margin of error is huge.

I used something basically like this as part of geotagging corners mathematically, but somehow the results weren't quite right and no amount of tweaking fixed things. Theoretically this should work and be easily automated, given that the camera sensor size is actually what the manufacture says it is.

My expertise lies in software engineering rather than photography and drone technology, so I may be going down the wrong path with regard to all of this, so I'm open to suggestions and curious as to general wisdom in this area. Setting a few ground control points in each image would probably do the trick, but is logistically prohibitive in a production environment so I'm trying to avoid that till all else.

 

[Droning on and on...]

Surely you're reinventing the wheel here?

Have you researched extensively what's already available to do this? Too expensive?

Are you extracting the logs, decoding them, and using this data? Position and height above takeoff data is in there and can easily be combined with topo data to estimate AGL more than good enough for this purpose post processing. The log even shows when you take a photo too, I think.

Here's an example of a logfile extracted post-flight. Since you're writing the code, see if the data here might be more useful than EXIF data.

 

[BassFromAbove]

mightypilot2000 yes I've looked at the flight logs - it doesn't really help as it doesn't seem to contain anything that gets me to either calculating the dimensions of the image footprint, corner coordinates or something equally as useful. The altitude data in my files' EXIFs seem to be actual AGL too.

That's what I thought re the wheel metaphor. Surely this has been done ad nauseum and a simple, low cost solution exists. Well if it does, it's not easily found regardless of simplicity or cost. I looked into Pix4d and various Esri offerings, namely their desktop stuff for proof-of-concept's sake regardless of the fact that in the end this needs to be automated and ideally integrated with a rather large internal application.

Both of these companies' desktop products imply or quasi-confirm that they will do at-least part or all of what I am aiming to achieve. In the end they are apparently designed to work with requirements and resources that are beyond the scope of my own. For example, Pix4d mapper requires at least three images to do anything. Do I have to take three needlessly overlapping images of a relatively tiny piece of land to use that? Is there something else required in the process that I've not yet bumped into that will require extra work? I've not yet jumped through the hoops to see if I can get a trial of Esri drone2map, which may actually be less restrictive than Pix4d mapper.

My project is in the prototype stage and I would like to see an example of output before paying upwards of $2k a year. These products are vast to the point where they make wonder why bother. Under the hood these most likely use the same algorithms as open-source GIS tools, and getting those tools setup to do what I want should be a matter of aligning the pipeline correctly and/or properly creating the inputs to the pipeline, but how? In addition, they are easily automated. I've already produced great results, the coordinates are just off a bit. Maddening.

 

[Droning on and on...]

mightypilot2000 yes I've looked at the flight logs - it doesn't really help as it doesn't seem to contain anything that gets me to either calculating the dimensions of the image footprint, corner coordinates or something equally as useful. The altitude data in my files' EXIFs seem to be actual AGL too.

A lot of us would like to know how AGL is possible. Could be it's just the altitude data from GPS and the barometer, available in the logs. Can you share an example from one of your shots? The AGL exif data, and your estimate of the elevation of the ground below?

If our drones are figuring out AGL somehow, there are a lot of people here that would love to know how!

Since altitude is recorded (and sensed somehow) in the logs, it is my speculation our drones reference GPS altitude at takeoff to calibrate the barometric sensor, then sense altitude from barometric changes from there.

 

[Meta4]

Hello all,

The DJI Mavic Air 2, or at least the one that I have which is about three years old now, does not include corner coordinate information in the EXIF part of the image files that it creates. I've seen corner coordinates in examples of other image files from other drone makes and models. .... Does anyone know if there is a way to get this drone to include corner coordinates, or have any recommendation of a similarly priced drone that does include corner coordinates?

Corner coordinates??
No DJI drone does this (and I'd be surprised if any other drone does).
There is no way to change the metadata that your camera records.

The altitude data in my files' EXIFs seem to be actual AGL too.

Your image metadata doesn't show actual AGL height.
Mainly because the drone has no way to measure AGL height.

Do I have to take three needlessly overlapping images of a relatively tiny piece of land to use that?

Overlapping for photogrammetry isn't "needless".

 

[PanoVolo]

I have never heard or seen corner coordinates in any drone metadata - xmp or exif. I suppose it can be calculated from position/altitude and camera fov. But keep in mind that altitude is not absolute, as you discovered, even the one that labeled as "absolute" in metadata. Also, reported camera focal length and fov in metadata is approximate, don't trust 24 mm focal length to be exactly that, it's rounded.

 

[Meta4]

I have never heard or seen corner coordinates in any drone metadata - xmp or exif. I suppose it can be calculated from position/altitude and camera fov.

It's correct that no drone gives corner coordinates
Since there is no way to measure the height of the drone above the ground, it can't be calculated.

But keep in mind that altitude is not absolute, as you discovered, even the one that labeled as "absolute" in metadata.

What DJI labels as Absolute Altitude (and GPS Altitude) is totally unreliable and can be out by >100 metres

Also, reported camera focal length and fov in metadata is approximate, don't trust 24 mm focal length to be exactly that, it's rounded.

DJI metadata does have the actual focal length of the lens as well as the 35mm equivalent.

 

[PanoVolo]

It's correct that no drone gives corner coordinates
Since there is no way to measure the height of the drone above the ground, it can't be calculated.

What DJI labels as Absolute Altitude (and GPS Altitude) is totally unreliable and can be out by >100 metres

DJI metadata does have the actual focal length of the lens as well as the 35mm equivalent.

it does have fov (not all drones) and 35 mm equivalent (all drones) - we use this info in PanoVolo ) but my point is these values are approximate, following CIPA standard. So 23.7 mm lens will be reported as 24 mm. It's ok fo rough guessing of fov, but is not accurate enough if you try to estimate where image corners are (agl problem no withstanding)

 

[PanoVolo]

it does have fov (not all drones) and 35 mm equivalent (all drones) - we use this info in PanoVolo ) but my point is these values are approximate, following CIPA standard. So 23.7 mm lens will be reported as 24 mm. It's ok fo rough guessing of fov, but is not accurate enough if you try to estimate where image corners are (agl problem no withstanding)

I wonder if enterprise drones with RTK can calculate corners though. Have no experience with them, so don't know for sure.