Helping PhotoScan with tricky aerial imagery

[maddin]

I have a somewhat difficult set of aerial photography, taken from a helicopter, unfortunately with suboptimal overlap (~ 30%) and with some images of forest/trees only. So, in order to get this set of images to a working state, I am looking for help on a few questions:

• I do not have GPS coordinates for the images but I do roughly know the order and layout of how the images were taken (the HC flew in a zig-zag path towards north) - is it possible to add this information to PhotoScan so it only tries to match neighboring images? In other words, I already know the rough layout/pattern of the images and PhotoScan should only 'refine' it.
• I read the manual and the online tutorials but still am not clear how I can manually add image features/virtual markers to help with difficult images?
• What can I do to manually help the alignment process? I am happy to delete erroneous points and/or matches between images but I am missing how to do this in a way that affects the alignment?
• Is it possible (and does it make sense) to start with a low accuracy and then subsequently refine the solution instead of redoing the photo alignment from scratch for each new accuracy setting?
• The images I work with are pretty high resolution (~70 megapixels each) - do I need to adjust some settings (e.g. max points per photo) in PhotoScan to correct for this?
• Alignment computation times for the entire set are rather long (> 3 hrs) - is there a smart way to break this down into smaller pieces that can be increased later bit by bit?

Thanks very much for any help and advice!
Martin

[Diego]

A basic principle of photogrammetry is the overlap, with 30%, you can not make link to photographs, between three or more photos, not even manually. End of story.

Photos with this overlap, only when you work with inertial measurement unit, GNSS, high precision, of course not your case. This is known as direct orientation.

You need to know the basic principles of photogrammetry, is the minimum before attempting any type of process.

[Wishgranter]

Maddin can share the photoset ? send me link / data to muzeumhb@gmail.com

BUT as Diegotorres say its need to know some "background" about photogrametry before doing some projects or you get in problems....

[maddin]

I understand that the overlap is problematic, but I did not take the pictures and got involved only afterwards. The basic questions still remain, though, even if this specific project might not benefit from it.

[maddin]

Any more insights on the original questions?

Thanks
Martin

[Wishgranter]

try adding POINTS even without GPS coordinates, place on same realworld spots, try so that add 1, link it to other images, then add 2nd link on other images and so on........ if possible, downsample few to 1/2 of resolution put in JPG and send it to me, will digg intro it....

[maddin]

try adding POINTS even without GPS coordinates, place on same realworld spots, try so that add 1, link it to other images, then add 2nd link on other images and so on........ if possible, downsample few to 1/2 of resolution put in JPG and send it to me, will digg intro it....

Thanks for the offer, but I am really interested in answers to my questions above, rather than finding someone to fix my project (which I would need to get extra permission to share).

Martin

[Diego]

Hope this solves your doubts

http://trackair.com/index.php/trackair-uav/

[Wishgranter]

Im preparing few thing about solving this..... thanx for the sample dataset....

[Alexey Pasumansky]

Hello Martin,

1) You can either use Generic pre-selection (Align Photos) or import fake-telemetry data in arbitrary local

coordinate system and use Ground Control pre-selection. It is not possible to introduce the order of the

images in any other way except Ground Control pane.

2) Markers could be used as valid matches between images. But this information is only used for "Align

selected cameras" feature and the following conditions should be fulfilled:
- N/A images (that you are going to align with the rest of successfully aligned dataset) should have at least

four markers placed on them,
- each marker from the previous point should be defined in 3D-space, thus having at least two projections

defined on the already aligned images.


3) Deleting obviously erroneous points from point cloud is reasonable for later camera alignment optimization. However, you should keep in mind that removing too much matches could affect alignment stability since true matches between images could be removed.

4) It is not possible to improve alignment accuracy. So you need to start over the alignment if you wish to use another accuracy settings.

5) We are not sure it is reasonable to increase "points per photo" parameter value even to high-res images. However, it is not prohibited. We suggest that experiments could be don in 40 000 - 100 000  points interval.

6) Align photos processing time mainly depend on the number of images, point per photo value, pre-selection, alignment accuracy and hardware resources.

[BretC]

Group: 

Does the X, Y, Z geocode values included on many geocoded photograph files enhance the method/speed in which Agisoft processes photographs? 

I am confused by the manual where it states "(camera orientation data, i.e. pitch, roll and yaw values, could also be imported, but the data is not obligatory)."  This goes against logic whereas the X,Y & Z data can be used to enhance AgiSoft's ability to position and align the photographs.

Thank you.
Bret C