Pair list. Speeding up pair matching

I used to experiment with VisualSfM and there was an option for the user to provide with a list of pairs so that the system does not have to guess wich images are overlapping others, or at least, in a limited extension.
It used to be of much improvement in my particular case that I am working with linear transects and am sure that image #001 has nothing to do with image #324 nor even with #020. I used to list 8 neighbours (4 preceding and 4 succeding) for each image si the amount of SIFT matching would be trmendously smaller knowing that I work with hundreds even thousands of frame extractions.
I would like to know if such possibility exists in webODM or ODM.
English is not my native language so I may have missed some rules or may have not been clear enough. Please, forgive me and feel free to let me know.

Your English is great, so no apologies there.

Do you have geographic coordinates associated with each of your images?

Well, thanks for your response,

That’s the problem I am trying to overcome.
We regularly tow an underwater video camera at 20 m depth and along a Km long transect to assess some variables there, and the problem is that we do not really know what the actual camera path is.
The environment I am trying to reproduce is the sea bottom along a transect, 20 m deep, where it is not easy to be exact about coordinates, unlike more shallow waters. It is the same problem as to mapping a cave from SfM, which I am also trying to solve with SfM from smartphones or other cheap cameras. I am not an advanced geek in terms of programming (as for now) but I definetly see I should be able to achieve the goal. I have made some progress with Meshroom, but I would like to give webODM a try.

I have been considering throwing concrete cubes with CGP painted on them throughout the area, but due to the fact that under the sea conditions (current drag, wind, etc.) its hard to really cover the same line as previous campaigns. If we could, CGP problem would be addressed once and for-ever. But in our case, getting the exact enough coordinates for a cube 20 m deep or more is kind of a challenge, let alone tens of them along the 6 transects we have to perform.
The only coordinate we can be sure enough is the starting point. We can also set some environmental friendly and durable CGP, but not much. As for now, I do not have any and would like to know how far I can go with this.
in VisualSfM it was possible to make a list of pairs inf you know ahead there is a reduced number of overlaps, which is very helpful.

Gotcha. You may have to dive a bit into OpenSfM to achieve this. @pierotofy may be able to point you where to start.

Thanks anyway, @smathermather-cm.
To give you an idea of how important is it, I have a running task still running since 5/4/2019 11:39:46, trying to match 665 pictures. Do the maths if you had to match every image against each other…
They are frames from a videotransect and none of them has to do with those that are 6 frames away. Best matches would be with the 4 preceding and 4 succeding, as I have said before. PNG I attatched a snapshot of what I mean.

1 Like