Best method for stitching aerial mapping images?   [SOLVED] - View the solution

Share your tips and tricks here or get help with any Autopano Pro / Giga problem!
No bug reports (of any kind) in this forum!
no avatar
rosscova
New member
 
Topic author
Posts: 2
Likes: 2 posts
Liked in: 2 posts
Joined: Mon Jan 09, 2017 7:50 am
Info

Best method for stitching aerial mapping images?

by rosscova » Mon Jan 09, 2017 8:03 am

Hi!

I'm looking to use AutoPano Giga for stitching aerial survey mapping images. The images are taken from a UAV flying with the camera pointed straight down. The UAV flies a grid pattern, taking images along the way, meaning that:

a) each images is taken from a different point of view, and
b) the overall panorama should be "flat" after processing (ie: images towards the edges shouldn't be stretched or contracted compared to those at the centre; all images should be represented the same*). It should represent a "map" of the surveyed area.

I thought the "multiple viewpoints" method would work well for this in AutoPano Giga, but I'm finding it doesn't do a good job of stitching the images (the normal mode works FAR better at finding appropriate control points and stitching the images together). So I'm hoping to stick with the regular detection method, but would like to know if there's a way to tell APG to render the image flat. I thought the "orthographic" projection would be appropriate, but I'm finding the "fisheye" projection with FOV set at 360 degrees seems to give a result closer to what I need.

Could anyone help with what would be the best workflow to get this kind of result in APG? Are people using this software for this purpose anywhere?

* note one reason for this is that it should be possible to roughly measure objects in the resulting image. To do this, it's important to make sure that each image is interpolated onto a flat plane, not curved.

no avatar
Nanard
Member
 
Posts: 150
Likes: 0 post
Liked in: 3 posts
Joined: Sun Jul 29, 2007 11:00 pm
Location: France
Info

Re: Best method for stitching aerial mapping images?

by Nanard » Mon Jan 09, 2017 10:38 am

Did you try to force a very long focal length, 1000 mm or more ?

rosscova likes this post.

no avatar
Panoram1x
Member
 
Posts: 236
Likes: 1 post
Liked in: 41 posts
Joined: Sun Jun 06, 2010 9:59 pm
Location: Barcelona (Catalunya)
Info

Re: Best method for stitching aerial mapping images?  [SOLVED]

by Panoram1x » Mon Jan 09, 2017 11:34 am

Surely your problem should look like this
viewtopic.php?f=2&t=32642

You could try these settings ...
viewtopic.php?f=2&t=31164#p181055

rosscova likes this post.

no avatar
rosscova
New member
 
Topic author
Posts: 2
Likes: 2 posts
Liked in: 2 posts
Joined: Mon Jan 09, 2017 7:50 am
Info

Re: Best method for stitching aerial mapping images?

by rosscova » Mon Jan 09, 2017 1:50 pm

Thank you @Panoram1x and @Nanard for the responses. I had thought of forcing a long focal length, but didn't realise it could be done within APG. I'll give it a try tomorrow, and let you know how I go.

I've got this working to where I'm very happy with the results, so I'll briefly outline the method I'm using:

1. Load all images into APG.
2. Under "group settings" > "optimization", tick "Optimization stages (manual)", and uncheck "Final lens distortion" *
3. Still in "group settings", go to the "render" tab > "blending presets" > "Advanced settings". I'm using "linear" blending, and "smart" cutting **. Then click OK.
4. Under "Image properties" change the focal length to a very large value (I've been using 20000mm with good results) ***, then click OK
5. Run "Detect", and work from there. (I go through and edit where necessary).
6. In selecting an appropriate projection, I'm finding that with such a long focal length selected, there are a few that work similarly well, but I've been settling on "Ortho" most of the time.

The results I'm getting now are great. I've been using Photoshop for these stitches until now, and (although the "Photomerge" function in Photoshop is very impressive) I can confidently say the APG out-performs Photoshop by a huge margin.

For anyone interested, I'm doing aerial mapping with a DJI Phantom 4, using DroneDeploy to control the capture phase. I'm feeding anywhere from 50 to around 250 images into APG per survey. I've got a reasonable computer to run this on (quad i7, 16GB RAM, Ubuntu 16.04), and would suggest the amount of RAM I'm using is the bare minimum for this kind of work. System Monitor shows around 80-90% RAM usage during processing, even with no other applications open.


* I decided to try leaving "First lens distortion" checked here, since I do want APG to correct for lens distortions, just not for the effects of camera rotation. My hypothesis here is that each image gets "corrected" on input (First lens distortion), but the panorama will not be "corrected" on output (Final lens distortion). That's the behaviour I want.

** I used "Anti-ghost" to begin with, but found that "simple" produces very good results, and seems to speed up the processing time significantly (a 6 hour job was reduced to around 1hr). I don't want images to be "blended" per-se, I want a sharp cut between them (similar to what Photoshop's "Auto-blend layers" does), so "simple" seems appropriate for me here.

*** My understanding here is that APG will use the focal length to figure out how much the camera rotated between exposures. For example, if a 50% overlap is found, this would correspond to a big rotation of a fish-eye lens (you'd must have turned your body a lot to get that overlap), but a very small rotation of a telephoto lens (you only turn very slightly to get the image overlapping that much). Since in my case, the actual camera rotation is zero (the camera position moved, as opposed to rotating on one spot), the larger the focal length APG thinks I used, the less it will try to account for a rotation (which never happened).

2 people like this post.


Who is online

Users browsing this forum: No registered users and 2 guests