Combining terrestrial (ground based) photos with aerial (uav) images

[Yoann Courtois]

Hi cadm8 !

Were your terrestrial pictures taken at another time than aerial pictures ? That may explain why they don't align together easily.

Markers are the handmade version of tie points (you need PhotoScan Pro).
They are then calling GCPs when you attach them 3D coordinates. But you are not required to !

So, the workflow would be:
- Create a main chunk with aerial pictures, align and georeference it.
- Create as many chunks as you have terrestrial sets of pictures (I mean one chunk per location) and align each one separately.
- Create a set of markers (3 mandatory, I would suggest you 4 in minimum) in each chunk on notable points. Be careful not to use same marker names in different chunks.
- Create those same markers in the main chunk and use same names so that one marker in a terrestrial chunk can be match with one marker in the main aerial chunk.
- Align all the chunks, based on marker, and keep the main one as reference. Check RMS errors in the console.
- If needed, merged all the chunks for further process.

Hope it's clear ^^
Regards

[cadm8]

Thanks for the reply. Painful strategy but seems right as an algorithm!

[Thomas_Steffen]

Hello, i don't know if the problem is solved already, but maybe you'll have a look at the 3D Image Vector: https://www.redcatch.at/de/products-mono/

I also use a Phantom 4 Pro, combined with a Sony a6500 with the 3D Image Vector. The big advantage of it is, that the orientation an position of the camera is recorded an can easily be imported in Photoscan. The accuracy of the camera positions is about 8cm, using Post-Processing. You nevertheless need GCP's, to align your photos from your Phantom 4 Pro and your terrestrial photos exactly.

[Yoann Courtois]

Hi Thomas !

Thanks for the help !
Indeed I guess the problem was solved since June :-)

Even if your device gives camera positions with a good accuracy (but still need to process the set of pictures), what about the camera orientation accuracy ?

As well, I don't use GCPs to align aerial pictures with terrestrial one anymore. Indeed with an organised way of taking terrestrial pictures, you can have enough overlap and so enough tie points between aerial and terrestrial pictures.
Still need known 3D points to get the global absolute orientation.

Regards

[zxt2320086]

Hi Thomas !

Thanks for the help !
Indeed I guess the problem was solved since June :-)

Even if your device gives camera positions with a good accuracy (but still need to process the set of pictures), what about the camera orientation accuracy ?

As well, I don't use GCPs to align aerial pictures with terrestrial one anymore. Indeed with an organised way of taking terrestrial pictures, you can have enough overlap and so enough tie points between aerial and terrestrial pictures.
Still need known 3D points to get the global absolute orientation.

Regards

Hi Yoann Courtois,

So glad you‘ve solved the problem.

any suggestion on the skill of taking terrestrial photos?Such as overlap percentage,or taking photos from ground or from a pole?

Regards

[Yoann Courtois]

Hi,

Firstly, I would say that aerial survey well match with pole survey, which well match with ground survey. Indeed, angle of view very differ from aerial to ground point of view, which is the weak point of SIFT algorithm. Pole survey is then a good compromise to insure the 3D reconstruction.

After that, is order to acquire data of an area invisible from aerial point of view (like several hundreds of m² in the middle of a 20ha drone survey), I would flight all above the area, and acquire pictures with a pole under the mask (trees ?), while overflow the masked area so that a strip aroung the mask area would be mapped from pole and drone point of view.
Overlap of both aerial and pole surveys could be keep normal, like 80% front and 70% side.

Finally, I would suggest you to use check points both in area exclusively mapped by drone and area mapped with the pole (and masked for the drone), so that you could improve you survey process and figure out the accuracy of your 3D model.

Regards

P.S. Feel free to share your results and some screen shots of your project