Hi all!
I very often find my self doing repeated missions on the same sites to detect changes throughout time (is birds incubating on nests in colonies in this case). I want to align the different layers so that I can easily see differences over time, and plot nests. Previously I have first processed each mission by it self in ODM before merging them with Photomerge in Photoshop (without blending layers). That have usually worked, with the obvious drawback that the photo is not georeferences anymore.
I would like to align the different missions to each other. They do not have to be precisely referenced in the real world. As I see it there are a few different options that all seems a bit difficult:
Use GCPs. It is however very impractical to access the sites and install GCPs before the flights, both since there are many sites (and flight times of between 2-10 minutes would rise a lot) and because the birds will be disturbed. Permanent GCPs are possible, but impractical in many cases as there are many sites that might move around between years.
Make ODM process all photos from one year at the same time and make it output multiple layers. As far as i know not possible.
Use computer vision to recognize features that are similar between flight and use them as “GCPs”, just with slightly inaccurate coordinates.
Find a GIS method to align the different layers based on content in the same way as Photoshop does, but keeping the coordinates.
Usually the missions are done flying free-hand. So the raw photos are rotated and shifted between different missions.
I hope some of you are able to point me in a good direction!
Both urban (roof-tops) and islands with either bushes or grass. There are usually quite alot of features like dead trees and similar one could use. It is however not that easy, as different missions usually also warp slightly. TIFFs of my attempt in qgis to manually align a couple can be found here drone.zip - Google Drive
Welcome! Awesome project! Also, love the avatar. My favorite bird is the Ring-Billed Seagull (though yours looks to be something different).
You hit upon a tough thing… GCPs will solve your surcey to survey reference problem, as well as preventing skew and other reconstruction errors in subsequent datasets. You also noted that they may not be possible to place…
Andreas brought up Manual Tie Points, which should serve a similar purpose in terms of keeping relative accuracy of reconstruction high, but identifying them reliably can be tough if they are not highly distinctive.
What would permanent ones look like for you? Are they tenable?
Thanks! It’s actually a pallas’s gull (first and only record for Norway)!
I’m painfully aware that propper gcps would fix everything I have used those a few times on other kinds of missions. That also means that I know how long it takes to manually plot all the gcps on the image files! With the scale we are talking about here it would take weeks to annotate, even with good chessboard gcps. If I had only one or a few colonies that would be perfectly acceptable, but I’m a bit hesitant with more than 5-6 missions (and usually a fixed amount of payed hours).
I’m really looking for an as automated process as possible. I’ve just started working with aruco gcps, and they seem to be the solution to many of these problems. Permanent just has to mean a few months, so it might be possible to just lay down printed/fabric aruco gcps.
Still, it would mean way less work, probably less disturbance to the colonies/sites, and maybe even quicker process times to have a photomerge with coordinates. Maybe one could photomerge, and then georefer the output? It does sound like a un-ideal work-around
