GCPs in the middle of nowhere

[George]

Hi all concerned with aerial photography,
just need your lovely advice ...
Imagine that you are in the middle of nowhere or say in the middle of no-no-no ... fcking nowhere ...
That may have much in common with Amazon river basin case ...
So fact is that it is literally impossible to collect any kind of GCPs.
You capture images of approx. 20-35 km2.
However, the problem still persists - you need fairly nice geo-registration; at least at the level of meters (definitely not dm or cm). Your on-board GPS reciever is a classic GPS which is far not of the high accuracy required for geo-registration (definitely not for Z coordinate).
What would be your method for GCPs collection and/or geo-registration?
Thanks in advance.

[RalfH]

If your camera and GPS are up in the air and you're not in a mountainous area, your GPS signal should be very good - you see many satellites and you don't have problems with signal reflection. So your standard GPS positions should be quite accurate (a few metres). If you collect many photographs over such a large area and use GPS positions of the camera (probably hundreds of them), the georeferencing should be quite accurate (errors get smoothed out). There are different options to further improve accuracy. You could use a GPS logger which uses a correction signal (e.g. WAAS or EGNOS, see http://en.wikipedia.org/wiki/GBAS#Ground-based_augmentation_system). Alternatively (or additionally), if you fly 2 GPS loggers, you can (a) create and compare 2 different DEM based on each logger's positions and (b) use the average positions of the 2 receivers to improve accuracy. If you have a receiver that can record "raw" GPS data (RINEX), you can post-process the data using PPP (there's an online service somewhere at Natural Resources Canada).

[George]

ok. What about Z accuracy then in relation to the actual terrain?
Actually that is the most worrying geo-registration I am after.

[RalfH]

I have just processed one project with 744 camera positions (standard GPS) for which the difference between camera GPS altitude was between -4 and +7 m. The difference between the two DSMs is between 1.5 and 2.3 m. One thing to keep in mind is that there are three basic types of GPS error:

(1) noise - the values just jump erratically about a position
(2) offset - the mean of the computed positions does not correspond exactly to the true position
(2) drift - the mean of the computed positions changes slowly over time

Of course, all of these errors occur at the same time.

Noise can easily be removed by throwing camera model coordinates and camera GPS coordinates into JavaGraticule3D. A stable offset is usually not very much of an issue. For example, if you are working on large areas you could correct for this using the SRTM evelation model. Drift is more tricky as it can cause your DEM to be tilted in space (which is really not what you want). Flying a lawnmower pattern and doing some extra flight lines perpendicular to this will reduce the impact of drift as opposed to a single line flight plan.

[George]

However, how do you then obtain correct Z value for your resulting DEM?
On-board GPS gives flight altitude only. Or ...?

[RalfH]

If you use the camera positions as "ground" control points, everything will work out, because the modelled camera positions are also above the modelled ground surface.

[George]

What would you suggest then for recalculation of altitudes of the cameras?
Or you need to know the ASL of the switch on or launch point ...

[George]

Well ... you know that the major question is then to be able mathematically and correctly compare the DSM produced and DTM ...

[RalfH]

Perhaps I don't quite understand your last two questions...

What would you suggest then for recalculation of altitudes of the cameras?
Or you need to know the ASL of the switch on or launch point ...

If you use the camera positions (modelled vs. GPS) as control points, your model will be georeferenced accordingly.

Well ... you know that the major question is then to be able mathematically and correctly compare the DSM produced and DTM ...

If this refers to comparing your Photoscan DSM with the SRTM global elevation model: SRTM uses short-wave radar which does not penetrate much into vegetation canopies. So SRTM is actually more a DSM than a DTM. The ASTER global DEM is based on photogrammetry using satellite imagery, so this too is actually a DSM.

[George]

This is really an interesting info! Thanks

Well ... you know that the major question is then to be able mathematically and correctly compare the DSM produced and DTM ...

If this refers to comparing your Photoscan DSM with the SRTM global elevation model: SRTM uses short-wave radar which does not penetrate much into vegetation canopies. So SRTM is actually more a DSM than a DTM. The ASTER global DEM is based on photogrammetry using satellite imagery, so this too is actually a DSM.
[/quote]

[George]

Using standard onboard GPS is pretty much ok for geo-referencing at XY, but what is then about Z value?
My question is more about how correctly define the Z value above sea level?
I suppose that then you need to know correctly (Z) at your switch-on position.

What would you suggest then for recalculation of altitudes of the cameras?
Or you need to know the ASL of the switch on or launch point ...

If you use the camera positions (modelled vs. GPS) as control points, your model will be georeferenced accordingly.

[RalfH]

Using standard onboard GPS is pretty much ok for geo-referencing at XY, but what is then about Z value?
My question is more about how correctly define the Z value above sea level?
I suppose that then you need to know correctly (Z) at your switch-on position.

Unfortunately, it is not that easy. Because of the drift problem (which is not predictable), knowing your starting position will not help. Even knowing your start and end position and interpolating the altitude correction inbetween will not work, because drift is not linear and can change direction. I have tried correcting GPS tracks using doppler data recorded in the raw GPS signal (works very well for smoothing out position noise), but there's also a drift in the doppler data, so it will not help you getting the true altitude. Also, using two standard receivers and trying to correct one based on the other will not work (only with differential GPS using the raw GPS signal), because alll errors can be very different in two identical devices sitting next to each other.

How accurately do you need your Z values anyway?

[George]

+-0,5 to 1 m at least would be great to have ...
in my case the principle is to have Z value correctly assigned ASL

[RalfH]

+-0,5 to 1 m at least would be great to have ...

That's not achievable with standard GPS. If you have a receiver that uses additional correction signals (EGNOS/WAAS/GBAS...) you could perhaps get into this range (using hundreds of individual GPS positions). Much better would be a receiver that records raw GPS data which you can post-process using PPP (http://www.geod.nrcan.gc.ca/products-produits/ppp_e.php) or differential GPS (with additional base station data), but suddenly we're talking about thousands of Euros...

[frank.stremke]

hello
maybe i can help out with some info here. what you need as i understand is basicly the correctet cammera position while taking photographs.
you can record gps raw data with a loger on the plane (record antenna deviaton from the camera position)
also during the same time have a other GPS rawdata loger placed on the ground closeby or in the survey area at a known location or make it a known location using long term measurement and post processing ppp. (way i just did in sudan)
then you process the recordings with RTKlib
http://www.rtklib.com/
depending on the aircraft speed and other factors such as signal quality obstructions ect. you can use 2 aproches for your data
1. Diferential gps, more robust
2. RTK Phase differential measurements more precise but the GPS signal needs to be recieveded continuisly to interuptions
depending on conditions and base line legth (distance between the two recievers) you can achive an accuracy of 0.5-2m for dgps or a few dm for rtk.
but since the aircraft moves (lets asume 120kts) 62m a second its all relative anyway. if you have your camera and gps clock syncronised its at least not a big problem as for altitude does not change so fast.
you can then process the RTKlib output data (textfile with 3d position) together with your pix in e.g. gpsprune
http://activityworkshop.net/software/gpsprune/
this will allow you to write the correctet GPS location into the exif data of your pictures
now you are ready to create your model. but you may have to georeference a bit due to the aircraft moving so fast.
iam pretty sure this is not a clean and accademic correct aproach but this is what i would do.
you will be able to by good hardware GPS reciever and hardware for this kind of work for around 1000euros complete build and testet or build it yourself using kits you will find plenty of information once you get into the RTKlib scene. but be advised its a steep learning curve.
btw i bought my hardware here:
https://www.optimalsystem.de/os.aspx?x=0
hope this helps
frank